CN109925002A - Artificial intelligence echocardiogram data collection system and its collecting method - Google Patents
Artificial intelligence echocardiogram data collection system and its collecting method Download PDFInfo
- Publication number
- CN109925002A CN109925002A CN201910034584.7A CN201910034584A CN109925002A CN 109925002 A CN109925002 A CN 109925002A CN 201910034584 A CN201910034584 A CN 201910034584A CN 109925002 A CN109925002 A CN 109925002A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- ultrasonic
- echocardiogram
- artificial intelligence
- 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
Abstract
A kind of artificial intelligence echocardiogram data collection system and the method using its progress data acquisition, which includes: data acquisition module (1), data uploading module (2), data identification module (3), ultrasonic imaging workstation (4), data measurement module (5) and data generation module (6).The system can assist doctor to carry out the accurate section positioning of echocardiogram and DATA REASONING, not only reduce ultrasonic cardiography results acquisition data deviation caused by artificial origin, moreover it is possible to substantially mitigate ultrasonic doctor work load.
Description
Technical field
The present invention relates to a kind of image data collection system more particularly to a kind of artificial intelligence technology is utilized to carry out the ultrasonic heart
The system of cardon data acquisition.
Background technique
Currently, artificial intelligence is just welcoming the innovation and development phase of a new round in the world, and basic reason is depth
Practise algorithm, Large-scale parallel computing and big data technology.Have in medical big data 90% from medical image, and China
Medical image data just with the dilatation year by year of 30% annual growth, is wanted to meet medical image analysis and the processing of more sophisticated
It asks, artificial intelligence approach becomes a research hotspot of medical image processing technology development in recent years.
As the important composition part of medical image, echocardiogram is to check angiocarpy using ultrasonic specific physical
A kind of woundless testing of system structure and function, main purpose are to evaluate the size of the dependency structures such as heart, form
And whether function is normal.Echocardiography is one of cardiovascular diseases screening, the indispensable inspection of diagnosis, in clinical practice work
Using more and more extensive in work, but also corresponding effective information can be provided for the treatment of many diseases, extended in one's hands
Art room, Cardiac Catheterization Room, for the detection and monitoring in being performed the operation and in interventional therapy.
It due to the indispensable of echocardiogram and is widely applied, the demands on examination of echocardiogram constantly increases in recent years
It is long.However, national ultrasonic doctor has 10 Wan Duoming, possess 10 years experiences still less than 30,000 people；Due to ultrasonic technique personnel training
Some cycles are needed, talent's output speed can not supply the gap problem of ultrasonic doctor in a short time, it is more difficult to support ultrasound
The growing demand for services of cardiogram.And since China has the hospital's distribution situation to specialize, medical resource is distributed not
Balance, 80% medical resource concentrate on 20% Big urban hospital, and the high-quality ultrasonic doctor of basic hospital is deficient, can not be accurate
Implement echocardiography, a large amount of patients will make echocardiogram and have to pour into Big urban hospital, lead to Big urban hospital
Burden overload, ultrasonic doctor work be excessively saturated.
Echocardiogram is measured and is diagnosed by the dynamic image of the different sections of doctor's acquisition mostly, this is to super
The technical level of sound doctor individual requires relatively high with diagnosis capability.By factors shadows such as conditions of patients, doctor's experience, site environments
It rings, identical doctor is for same position patient or different doctors for the echocardiogram measurement consistency of same position patient and repetition
Property is difficult to realize, to fail to pinpoint a disease in diagnosis, mistaken diagnosis, delays the treatment of patient.Therefore, it is badly in need of one kind and can reduce human error causing
The method of ultrasonic cardiography results acquisition data deviation had not only facilitated the foundation of echocardiography professional standard, but also can delay
The problem of solving base's ultrasonic doctor shortage of resources.
In view of the above basis, a kind of echocardiogram data collection system based on artificial intelligence technology is herein proposed, it can
To assist doctor to carry out the accurate section positioning of echocardiogram and DATA REASONING, ultrasound caused by artificial origin is not only substantially reduced
Results acquisition data deviation aroused in interest, can also mitigate ultrasonic doctor work load, the work for keeping its more convenient, more efficient.
Summary of the invention
It is an object of the invention to: conventional ultrasound cardiogram is blended with artificial intelligence, exploitation unified standardization surpasses
Sound data measuring and acquisition system fills up industry blank.Acquisition system is measured by artificial intelligence ultrasound data, doctor can be assisted to know
Common section in other ultrasonic cardiography inspection, realizes to the positioning of key point/area and measurement in section, can not only automatically generate
Ultrasonic master data report, can also realize the intelligent prompt to abnormal ultrasonic diagnostic result.
The problem of present invention can solve includes: the unified standard for promoting ultrasonic examination by artificial intelligence technology, not only
The accuracy for promoting echocardiography, can solve the problem of Different hospital ultrasonic cardiography diagnostic result is not recognized each other, and promote
The intelligent development of echocardiography work, optimizes Allocation of Medical Resources, reduces medical treatment cost；It can also maximize and release
Ultrasonic doctor resource is put, and base doctor is made to have the ultrasonic cardiography checking ability of expert's grade, helps to realize patient point
Stream is medical, and the status of domestic ultrasonic medical maldistribution of the resources is effectively relieved.
The present invention provides a kind of artificial intelligence echocardiogram data collection system, comprising: on data acquisition module, data
Transmission module, data identification module, ultrasonic imaging workstation, data measurement module and data generation module；Wherein, data acquire
Module is used to check that data are acquired to ultrasonic cardiography；Data uploading module for will check that data upload, and on
Pass to data identification module；Data identification module quickly identifies data using convolutional neural networks；Data measurement module
For measuring required data；Data generation module automatically generates ultrasonic master data report according to the data that data measurement module measures
It accuses, and sends ultrasonic imaging workstation for report.It is adopted the present invention also provides a kind of using artificial intelligence ultrasonic cardiography diagram data
The method that collecting system carries out data acquisition, described method includes following steps:
Step 1 is acquired using data of the data acquisition module to patient's ultrasonic cardiography inspection, then by patient into
The data of row ultrasonic cardiography inspection upload to data identification module by data uploading module；
Step 2 is quickly identified using convolutional neural networks propagated forward algorithm by data identification module, judges the heart
Whether dirty ultrasonic cardiography section data comply with standard, if data are exported to ultrasonic imaging workstation by data fit standard；
If data are not inconsistent standardization, repeatedly step 1；
Step 3, after determining data fit standard and data being exported to ultrasonic imaging workstation, data measurement module is opened
Then data needed for beginning automatic measurement automatically generate ultrasonic base according to the data that data measurement module measures by data generation module
Notebook data report, and ultrasonic imaging workstation is sent by report, doctor can carry out the master data report automatically generated
Editor uses.
Artificial intelligence echocardiogram data collection system of the invention has following technical effect outstanding:
1. artificial intelligence technology, which is applied to ultrasonic cardiography, for the first time measures evaluation work, echocardiogram basic number is realized
According to the unified standard of collecting flowchart, industry blank is filled up.
2. developed using entirely autonomous original artificial intelligence system (zANN), belong to China first possess it is complete from
The artificial intelligence echocardiogram data collection system of main intellectual property.
3. artificial intelligence ultrasound data of the invention measures acquisition system, section is commonly used in achievable ultrasonic cardiography inspection
Automatic identification；It realizes to key point/area intelligent positioning and automatic measurement in section；Automatically generate ultrasonic master data report；
With realization to the intelligent prompt of abnormal ultrasonic diagnostic result.
4. system operatio of the invention is simply easily grasped, the workflow of hardly existing ultrasonic department, change hospital is universal
Threshold is low, can be widely used for various ultrasonic devices and medical institutions at different levels.
Detailed description of the invention
Fig. 1 is to carry out data acquisition using the artificial intelligence echocardiogram data collection system of the preferred embodiment of the present invention
Step schematic diagram；
Fig. 2 is the work flow diagram of the artificial intelligence echocardiogram data collection system of the preferred embodiment of the present invention.
Fig. 3 is the standard network model schematic of CNN.
Fig. 4 is the testing process schematic diagram of RCNN.
Specific embodiment
Artificial intelligence technology is applied to echocardiogram by artificial intelligence echocardiogram data collection system of the invention
Data acquisition in, can be realized the unified standard of echocardiogram data acquisition flow, reduce a large amount of human error and produce
It is raw.
Artificial intelligence echocardiogram data collection system of the invention has mainly used following several core technologies:
1, Preprocessing Technique
Image preprocessing is to come out each character image sorting to give identification module identification, this process is known as scheming
As pretreatment.In image analysis, the processing carried out before feature extraction, segmentation and matching is carried out to input picture.
The main purpose of image preprocessing is to eliminate information unrelated in image, restores useful real information, enhancing has
The detectability and simplified data to the maximum extent of information are closed, so that improves feature extraction, image segmentation, matching and identification can
By property.The conventional steps of image preprocessing are as follows:
1) image filtering.The energy of signal or image is largely focused on the low frequency and Mid Frequency of amplitude spectrum, and higher
Frequency range, useful information are often flooded by noise, it is therefore desirable to target figure under conditions of retaining image minutia as far as possible
The noise of picture is inhibited, and the quality for the treatment of effect will directly influence the validity of subsequent image analysis and identification and reliable
Property.The image filtering mode supported at present has: mean filter, gaussian filtering, median filtering and bilateral filtering；
2) image pyramid and size scaling.Image pyramid is one kind of multi-scale expression in image, is one kind with more
Resolution ratio carrys out the simple structure of effective and concept of interpretation of images.Before carrying out feature extraction and identification to image, usually need
Size change over is carried out to image in order to obtain the characteristics of image of different scale, in convolutional neural networks (Convolutional
Neural Network, referred to as: CNN) rise after, the extraction of image different scale feature has incorporated the design of network structure
In, therefore the pretreatment of this step mainly carries out size scaling to image, the image data of unified resolution is obtained, in order to subsequent
CNN network training；
3) Image Reversal.Image classification identification task in, the overturning of image should not influence identify as a result, therefore
When training CNN network, random overturning training image is needed, the model that training obtains in this way has translation invariance.Branch at present
The Image Reversal pretreatment held is spun upside down, left and right is overturn, 90 ° of overturnings；
4) image color adjustment and normalization.It is similar with Image Reversal, adjust the brightness of image, contrast, saturation degree and
Color equivalent parameters shall not influence the final effect of image recognition, therefore in training pattern, adjustment image that can be random
These attributes, to make the obtained model of training is as few as possible to be influenced by irrelevant factor.Finally carry out the normalization of image
Operation refers to that it is 0 that graphics standard, which is turned to pixel intensity mean value, and the image that variance is 1, the model trained in this way has illumination
Invariance.
According to the demand of different identification missions, one or more combinations of above-mentioned image preprocessing can be arbitrarily selected, are changed
Into the stability and reliability of subsequent image feature extraction and identification.
In the acquisition of ultrasonic image data, it is difficult as common training mission, there are 100,000 or even million or more quantity
The data of grade, at this time Preprocessing Technique can carry out image adding the operation such as make an uproar, invert, enhance, and data volume is double
Picture quality can also be improved.Then ready data are sent into convolutional neural networks (CNN) and are trained, so as to network science
To richer knowledge and the fault-tolerance of increase model.
2, convolutional neural networks (Convolutional Neural Network, referred to as: CNN)
Convolutional neural networks (CNN) are grown up on the basis of multilayer neural network for image classification and identification
And a kind of specially designed deep learning method.It is the machine learning model under a kind of supervised learning of depth, is had extremely strong
Adaptability, be good at mining data local feature, extract global training characteristics and classification, weight shared structure network are allowed to more
Similar to biological neural network, good achievement is all achieved in pattern-recognition every field.
CNN has the advantages that some traditional technologies are unexistent: good fault-tolerant ability, parallel processing capability and self study energy
Power, can processing environment information it is complicated, background knowledge is unclear, the problem of in the indefinite situation of inference rule, allow sample have compared with
Big defect, distortion, the speed of service is fast, and adaptive performance is good, resolution ratio with higher.Meanwhile generalization ability is significant
Better than other methods, convolutional neural networks have been applied to pattern classification, object detection and object identification etc..Utilize convolution
Neural network pattern classifier is directly used in gray level image using convolutional neural networks as general pattern classifier.
The image library of 10 common sections in echocardiogram is established in the present invention using CNN, and target image is carried out
Artificial mark identifies that AI model training provides required data for section.The figure based on CNN is established according to fixed identification target
As classifier, Boot Model training, and introduces appraisement system and generate evaluation parameter.When evaluation parameter reaches desired value,
Identification model training in echocardiogram section is completed.Specific step is as follows:
1) the image slices library of common 10 sections of echocardiogram is established；
2) image in image slices library is manually marked, i.e. the corresponding image tag of each picture, label
Totally ten class；
3) it by picture and its corresponding label, inputs CNN network and is trained, by hundreds of thousands of iteration, optimize CNN
The standard network model of the parameter of network, CNN is as shown in Figure 3: construct CNN Primary layer altogether there are three types of: convolutional layer, pond layer
With full articulamentum, the different local feature of image is mainly extracted in the effect of convolutional layer, and pond layer completes Fusion Features and dimensionality reduction,
Type inference and output are finally made by full articulamentum, wherein the parameter of convolutional layer and full articulamentum needs constantly training to optimize；
4) model of generation is tested, when the predictablity rate of model is higher than desired value, model training is completed, no
Then continue training until reaching requirement, finally obtains the CNN network model for adapting to different images section.
Pretreated data will be passed through, according to the packed data of certain format (Data) and label (Label), be sent into CNN
Network carries out supervised learning, continues to optimize Model Weight by back-propagation algorithm (Back propagation), reaches finger
Present weight hyper parameter, i.e. network model are saved after fixed loss (Loss).The image detected will be needed to be input to the net again
Network model carries out the propagated forward operation (Forward propagation) of CNN network, output using the weighted value kept
The testing result of the image.
3, Medical Image Segmentation Techniques
Medical image segmentation is the complexity of Medical Image Processing and analysis field and the step of key, and the purpose is to by medicine
Partial segmentation in image with certain particular meanings comes out, and extracts correlated characteristic, mentions for clinic diagnosis and pathological research
For reliable foundation, doctor is assisted to make more accurate diagnosis.
But information is complicated in medical image, intensity profile is uneven, and noise is larger, and shape easily occurs for organ-tissue
Become, these factors increase the difficulty of Feature Selection and image segmentation, as a consequence it is hardly possible to using a general method to all
Medical image realizes segmentation.Previous Medical Image Segmentation Techniques mostly use the conventional segmentation methods such as threshold method, present invention innovation
Using a series of image partition methods based on deep learning, image segmentation speed is substantially increased, and it is super to divide accuracy rate
Conventional segmentation methods are crossed, do not need manually to extract characteristics of image or image is excessively pre-processed, thus solve medicine figure
As being difficult to the difficulties divided.
Traditional image detecting method is mainly region suggestion+craft characteristic Design+classifier, the method that this system uses
It is returned for selective search method+CNN feature+Softmax classification+circumscribed rectangle, i.e. RCNN.
It is compared with conventional method, the image feature extraction procedure of RCNN transfers to CNN network to complete, it is no longer necessary to manually set
Meter, the performance of detection are classified Average Accuracy (mean Average Precision) more than 40 percentages higher than conventional method more
Than.Specific step is as follows for RCNN network training (referring to fig. 4):
1) selective search method.Objects in images region that may be present should have certain similitudes or continuity
Region.Therefore, selection search extracts bounding using the method that subregion merges based on this idea above
Boxes (circumscribed rectangle) boundary candidate frame.Firstly, being split algorithm to input picture generates many small subregions.Secondly,
Region merging technique is carried out according to similitude between these subregions (similarity standard mainly has color, texture, size etc.), constantly
The iteration merging of carry out region.Bounding boxes (circumscribed square is done to these subregions merged in each iterative process
Shape), the candidate frame that the circumscribed rectangle of these subregions is just known as；
2) CNN image characteristics extraction.It is had been described in front of this step technique, it is herein it should be noted that traditional
CNN network is mainly used for target classification, and the requirement of target classification is with space-invariance, therefore the layer of CNN convolutional network
Number is deeper, and the spatial information of target is lost more, but target detection/segmentation will obtain the accurate location information of target, because
This RCNN network introduces position sensing pond (Position-sensitive RoI pooling)；
3) Softmax classification+circumscribed rectangle returns.Softmax classification layer is similar with traditional CNN, is a full connection
Layer, output node number are the target classification number that network can identify, the probability of each classification are provided to input candidate frame, together
Shi Shengcheng Classification Loss value.Unlike classification task, it is for selecting the first step that more herein circumscribed rectangles, which return,
The candidate frame position that selecting property searching method provides is corrected, while generating position penalty values, and two penalty values are added to obtain
Final Detection task penalty values；
4) training is iterated to network, and the model of generation is tested, when mAP (more average standards of classification of model
True rate) deconditioning when meeting the requirements.
Image Segmentation Technology aims at each of tag image pixel, and each pixel is indicated with it
Classification is mapped.Because each of meeting forecast image pixel, is generally known as dense prediction for such task.No
For the region area of rule with common Bounding Box (bounding-box) come inaccuracy at last, image Segmentation Technology can will be ultrasonic
Irregular area area in image is split, in the hope of accurately area.
Fig. 1 is to carry out data acquisition using the artificial intelligence echocardiogram data collection system of the preferred embodiment of the present invention
The step schematic diagram of work.
Artificial intelligence echocardiogram data collection system of the invention includes following several modules: data acquisition module 1,
Data uploading module 2, data identification module 3, ultrasonic imaging workstation 4, data measurement module 5 and data generation module 6.
Wherein, data acquisition module 1 is used to check that data are acquired to ultrasonic cardiography；Data uploading module 2 will be for that will check data
It is uploaded, and uploads to data identification module 3；Data identification module 3 carries out data using convolutional neural networks (CNN)
Quickly identification；Data measurement module 5 is for measuring required data；The number that data generation module 6 is measured according to data measurement module 5
It is reported according to ultrasonic master data is automatically generated, and sends ultrasonic imaging workstation 4 for report.Data acquisition module (1) includes
Ultrasonic probe；Ultrasonic imaging workstation (4) includes terminal handler and display.
From figure 1 it appears that the step of preferably carrying out data acquisition using data collection system of the invention includes:
Step 1 is acquired using data of the data acquisition module 1 to patient's ultrasonic cardiography inspection, then by patient into
The data of row ultrasonic cardiography inspection upload to data identification module 3 by data uploading module 2；
Step 2 utilizes convolutional neural networks (CNN) propagated forward algorithm (Forward by data identification module 3
Propagation it) is quickly identified.The wherein training that the exploitation of data identification module 3 is iterated by data, then reach
When ideal loss (Loss) value, its weight parameter is saved, exports data identification module 3.Judge cardiac ultrasonic section number aroused in interest
According to whether complying with standard, if data fit standard, data are exported into ultrasonic imaging workstation 4；If data are not met
Standard, then repeatedly step 1；
Judge whether cardiac ultrasonic section aroused in interest complies with standard: if cardiac ultrasonic data fit standard in section aroused in interest, number
The prompt tone of " section standard " can be issued according to identification module 3, and the prompt of section standard is fed back to ultrasonic imaging workstation 4；
If cardiac ultrasonic section data aroused in interest are not inconsistent standardization, data identification module 3 can be sounded an alarm, and correct operator couple
The probe orientation of ultrasound cardiograph is adjusted, until the probe of ultrasound cardiograph gets ideal standard section, then, data
Identification module 3 feeds back the prompt of section standard to ultrasonic imaging workstation 4.
Step 3, after determining data fit standard and data being exported to ultrasonic imaging workstation 4, data measurement module 5
Data needed for starting automatic measurement, are then automatically generated by data generation module 6 according to the data that data measurement module 5 measures
Ultrasonic master data report, and ultrasonic imaging workstation 4 is sent by report, doctor can be to the master data report automatically generated
Announcement carries out editor's use.
Through the above steps, artificial intelligence echocardiogram data collection system of the invention may be implemented following several
Kind function:
1, to the common intelligent recognition for checking section of heart
Data identification module 3 often respectively can know common section with checks sequence according to echocardiogram automatically
Not, the left room long axis of parasternal, parasternal main artery short axle, the left room short axis view of parasternal, the apical four-chamber heart can accurately be identified
Section, the anxious face of five chamber of the apex of the heart, apex of the heart cor biloculare section, Four-chamber view under xiphoid-process, suprasternal fossa arch of aorta section etc..
2, the positioning of intelligently guiding heart sections
When heart ultrasonic cardiography section, data are not inconsistent standardization, data identification module 3 can intelligently guiding operator couple
Probe orientation is adjusted, until probe gets ideal standard section.
3, data acquisition is automatically performed in target section
Data measurement module 5 is automatically performed the acquisition of the measurement to echocardiogram common counter based on section, such as: rising actively
Arteries and veins internal diameter, left room Shu Monei warp, the last internal diameter of left room receipts, ejection fraction, chamber interval thickness, interventricular septum motion amplitude, left atrial size,
Right Fang great little, pulmonary artery trunk diameter, right ventricle's anteroposterior diameter, E wave maximum flow rate, A wave maximum flow rate etc..
4, the acquisition to cardiac structural, valvular data is completed
Data measurement module 5 observes the continuous implementations of ventricular atrial, and whether there is or not local motion exceptions, observes valve number shape
State is opened and closed situation, measures across version pressure difference, maximum flow rate, valve orifice area.Whether there is or not valvular regurgitations etc. for observation.
Fig. 2 shows the work flow diagrams of artificial intelligence echocardiogram data collection system, it can be seen from the figure that this
It is as follows that the artificial intelligence echocardiogram data collection system of invention preferred embodiment receives the operation carried out after measurement image:
1) firstly, being acquired using data of the data acquisition module 1 to patient's ultrasonic cardiography inspection, and pass through data
The echocardiogram of original measurement is uploaded to data identification module 3 by uploading module 2；
2) then, data identification module 3 carries out noise reduction to the echocardiogram being originally inputted using Preprocessing Technique
With image feature region intensive treatment；And the segmentation of image-region is carried out using the neural network model of region detection；It will divide
The image cut is sent into the convolutional neural networks model kept, by the propagated forward algorithm of neural network model, to output
Result make intelligent decision, judge whether the cardiac ultrasonic got section aroused in interest complies with standard；
3) after the cardiac ultrasonic got section aroused in interest complies with standard, by 5 pairs of data measurement module segmentation after image into
Row picture recognition and DATA REASONING, including target section positioning, region recognition positioning and cutting region one-point measurement, then by
Data generation module 6 automatically generates ultrasonic master data report.The data measurement module 5 is instructed using convolutional neural networks (CNN)
Practise the network model of zonule precise measurement.
By above-mentioned operation, the accuracy of echocardiography can be promoted, is effectively reduced caused by manual operation accidentally
Difference, while it can be obviously improved the working efficiency of doctor, keep its ultrasonic examination work more efficient, more convenient, realizes ultrasonic cardiography
The unified standard of figure master data collecting flowchart helps to solve asking of not recognizing each other of Different hospital ultrasonic cardiography diagnostic result
Topic reduces medical treatment cost.
Further, artificial intelligence echocardiogram data collection system easy grasp easy to operate of the invention, hardly
Change the workflow of existing ultrasonic department, hospital, Clinical practicability is strong, and promotional value is high, can assign base doctor " expert's grade "
Ultrasonic cardiography checking ability, the insufficient status of primary care service ability can be improved to a certain extent, be widely portable to
Various ultrasonic devices and medical institutions at different levels.
Claims (10)
1. a kind of artificial intelligence echocardiogram data collection system, comprising: data acquisition module (1), data uploading module
(2), data identification module (3), ultrasonic imaging workstation (4), data measurement module (5) and data generation module (6)；Its
In, data acquisition module (1) is used to check that data are acquired to ultrasonic cardiography；Data uploading module (2) will be for that will check number
According to being uploaded, and upload to data identification module (3)；Data identification module (3) using convolutional neural networks to data into
Row quickly identification；Data measurement module (5) is for measuring required data；Data generation module (6) is according to data measurement module (5)
The data of measurement automatically generate ultrasonic master data report, and send ultrasonic imaging workstation (4) for report.
2. a kind of utilize artificial intelligence echocardiogram data collection system as described in claim 1, wherein data identify mould
Block (3) carries out noise reduction and image feature region intensive treatment to the echocardiogram being originally inputted using Preprocessing Technique,
And the segmentation of image-region is carried out using the neural network model of region detection；And the image feeding after segmentation is kept
Convolutional neural networks model intelligent decision is carried out to the result of output, is sentenced by the propagated forward algorithm of neural network model
Whether the disconnected cardiac ultrasonic section data aroused in interest got comply with standard.
3. a kind of utilize artificial intelligence echocardiogram data collection system as described in claim 1, wherein data acquisition module
Block (1) includes ultrasonic probe；Ultrasonic imaging workstation (4) includes terminal handler and display.
4. a kind of side for carrying out data acquisition using artificial intelligence echocardiogram data collection system as described in claim 1
Method, described method includes following steps:
Step 1 is acquired using data of the data acquisition module (1) to patient's ultrasonic cardiography inspection, then carries out patient
The data of ultrasonic cardiography inspection upload to data identification module (3) by data uploading module (2)；
Step 2 is quickly identified using convolutional neural networks propagated forward algorithm by data identification module (3), judges heart
Whether ultrasonic cardiography section data comply with standard, if data are exported to ultrasonic imaging workstation by data fit standard
(4)；If data are not inconsistent standardization, repeatedly step 1；
Step 3, after determining data fit standard and data being exported to ultrasonic imaging workstation (4), data measurement module (5)
Data needed for starting automatic measurement, are then given birth to by data generation module (6) according to the data that data measurement module (5) measures automatically
It is reported at ultrasonic master data, and sends report to ultrasonic imaging workstation (4), doctor can be to the basic number automatically generated
It was reported that carrying out editor's use.
5. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
Wherein step 2 further include: if cardiac ultrasonic data fit standard in section aroused in interest, data identification module (3) can issue " section
The prompt tone of standard ", and the prompt of section standard is fed back to ultrasonic imaging workstation (4)；If cardiac ultrasonic section aroused in interest
Data are not inconsistent standardization, and data identification module (3) can sound an alarm, and correct operator to the probe side of ultrasound cardiograph
To being adjusted, the data of patient's ultrasonic cardiography inspection are resurveyed to re-use data acquisition module (1), directly
Probe to ultrasound cardiograph gets ideal standard section.
6. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
Wherein data identification module (3) carries out noise reduction and image feature to the echocardiogram being originally inputted using Preprocessing Technique
Region intensive treatment；And the segmentation of image-region is carried out using the neural network model of region detection；The image of segmentation is sent
Enter the convolutional neural networks model kept.
7. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
Wherein the process quickly identified using data identification module (3) in step 2 includes: normal according to echocardiogram
Automatic identification is carried out to common section respectively with checks sequence, accurately identifies that the left room long axis of parasternal, parasternal main artery are short
The left room short axis view of axis, parasternal, the anxious face of apical four-chamber, the anxious face of five chamber of the apex of the heart, apex of the heart cor biloculare section, four chambers under xiphoid-process
Anxious face, suprasternal fossa arch of aorta section etc..
8. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
Wherein data measurement module (5) is automatically performed the measurement to echocardiogram common counter based on cardiac ultrasonic section data aroused in interest
Acquisition, these data include: aorta ascendens internal diameter, left room Shu Monei warp, the last internal diameter of left room receipts, ejection fraction, chamber interval thickness,
Interventricular septum motion amplitude, left atrial size, right Fang great little, pulmonary artery trunk diameter, right ventricle's anteroposterior diameter, E wave maximum flow rate, A wave max-flow
Speed etc..And the continuous implementations of data observation ventricular atrial of the data measurement module (5) based on measurement, it is different that whether there is or not local motions
Often, valve number form, opening and closing situation are observed；Measure across version pressure difference, maximum flow rate, valve orifice area；Whether there is or not valvular regurgitations for observation.
9. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
After the cardiac ultrasonic got data fit standard in section aroused in interest, figure is carried out to the image after segmentation by data measurement module (5)
Piece identification and DATA REASONING, the one-point measurement including the positioning of target section, region recognition positioning and cutting region, the DATA REASONING
Module (5) trains the network model of zonule precise measurement using convolutional neural networks.
10. the method as claimed in claim 4 for carrying out data acquisition using artificial intelligence echocardiogram data collection system,
Wherein data acquisition module (1) includes ultrasonic probe；Ultrasonic imaging workstation (4) includes terminal handler and display.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910034584.7A CN109925002A (en) | 2019-01-15 | 2019-01-15 | Artificial intelligence echocardiogram data collection system and its collecting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910034584.7A CN109925002A (en) | 2019-01-15 | 2019-01-15 | Artificial intelligence echocardiogram data collection system and its collecting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109925002A true CN109925002A (en) | 2019-06-25 |
Family
ID=66985042
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910034584.7A Pending CN109925002A (en) | 2019-01-15 | 2019-01-15 | Artificial intelligence echocardiogram data collection system and its collecting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109925002A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110570409A (en) * | 2019-09-05 | 2019-12-13 | 李胜利 | method for acquiring fetal four-chamber cardiac section cardiac cycle video based on hybrid convolutional network |
| CN110742654A (en) * | 2019-11-05 | 2020-02-04 | 深圳度影医疗科技有限公司 | Method for positioning and measuring standard tangent plane based on three-dimensional ultrasonic image |
| CN110742653A (en) * | 2019-10-31 | 2020-02-04 | 无锡祥生医疗科技股份有限公司 | Cardiac cycle determination method and ultrasonic equipment |
| CN111904469A (en) * | 2020-08-07 | 2020-11-10 | 长沙大端信息科技有限公司 | Heart section detection method and system capable of realizing parallel processing |
| CN111915562A (en) * | 2020-07-02 | 2020-11-10 | 杭州深睿博联科技有限公司 | Deep learning children echocardiogram standard tangent plane identification method and device |
| CN112914610A (en) * | 2021-01-22 | 2021-06-08 | 华中科技大学同济医学院附属同济医院 | Contrast-enhanced echocardiography wall thickness automatic analysis system and method based on deep learning |
| CN114366165A (en) * | 2022-01-25 | 2022-04-19 | 上海杏脉信息科技有限公司 | Heart ultrasonic scanning auxiliary system, medium and electronic equipment |
| CN115760851A (en) * | 2023-01-06 | 2023-03-07 | 首都儿科研究所附属儿童医院 | Ultrasonic image data processing method and system based on machine learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680481A (en) * | 2013-11-28 | 2015-06-03 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic auxiliary scanning method and ultrasonic auxiliary scanning system |
| CN105555198A (en) * | 2014-03-20 | 2016-05-04 | 深圳迈瑞生物医疗电子股份有限公司 | Method and device for automatic identification of measurement item, and ultrasound imaging apparatus |
| CN106230891A (en) * | 2016-07-18 | 2016-12-14 | 上海市第人民医院 | A kind of method and system transmitting ultrasoundcardiogram report |
| WO2017206023A1 (en) * | 2016-05-30 | 2017-12-07 | 深圳迈瑞生物医疗电子股份有限公司 | Cardiac volume identification analysis system and method |
| EP3376472A1 (en) * | 2017-03-13 | 2018-09-19 | Koninklijke Philips N.V. | Anatomical measurements from ultrasound data |
| CN108734694A (en) * | 2018-04-09 | 2018-11-02 | 华南农业大学 | Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn |
-
2019
- 2019-01-15 CN CN201910034584.7A patent/CN109925002A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680481A (en) * | 2013-11-28 | 2015-06-03 | 深圳迈瑞生物医疗电子股份有限公司 | Ultrasonic auxiliary scanning method and ultrasonic auxiliary scanning system |
| CN105555198A (en) * | 2014-03-20 | 2016-05-04 | 深圳迈瑞生物医疗电子股份有限公司 | Method and device for automatic identification of measurement item, and ultrasound imaging apparatus |
| WO2017206023A1 (en) * | 2016-05-30 | 2017-12-07 | 深圳迈瑞生物医疗电子股份有限公司 | Cardiac volume identification analysis system and method |
| CN106230891A (en) * | 2016-07-18 | 2016-12-14 | 上海市第人民医院 | A kind of method and system transmitting ultrasoundcardiogram report |
| EP3376472A1 (en) * | 2017-03-13 | 2018-09-19 | Koninklijke Philips N.V. | Anatomical measurements from ultrasound data |
| CN108734694A (en) * | 2018-04-09 | 2018-11-02 | 华南农业大学 | Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn |
Non-Patent Citations (1)
| Title |
|---|
| 陶攀、付忠良、朱锴等: "基于深度学习的超声心动图切面识别方法", 《计算机应用》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110570409A (en) * | 2019-09-05 | 2019-12-13 | 李胜利 | method for acquiring fetal four-chamber cardiac section cardiac cycle video based on hybrid convolutional network |
| CN110570409B (en) * | 2019-09-05 | 2022-02-18 | 南方医科大学 | Method for acquiring fetal four-chamber cardiac section cardiac cycle video based on hybrid convolutional network |
| CN110742653A (en) * | 2019-10-31 | 2020-02-04 | 无锡祥生医疗科技股份有限公司 | Cardiac cycle determination method and ultrasonic equipment |
| WO2021082272A1 (en) * | 2019-10-31 | 2021-05-06 | 无锡祥生医疗科技股份有限公司 | Method for determining cardiac cycles and ultrasonic device |
| CN110742654A (en) * | 2019-11-05 | 2020-02-04 | 深圳度影医疗科技有限公司 | Method for positioning and measuring standard tangent plane based on three-dimensional ultrasonic image |
| CN111915562A (en) * | 2020-07-02 | 2020-11-10 | 杭州深睿博联科技有限公司 | Deep learning children echocardiogram standard tangent plane identification method and device |
| CN111904469A (en) * | 2020-08-07 | 2020-11-10 | 长沙大端信息科技有限公司 | Heart section detection method and system capable of realizing parallel processing |
| CN112914610A (en) * | 2021-01-22 | 2021-06-08 | 华中科技大学同济医学院附属同济医院 | Contrast-enhanced echocardiography wall thickness automatic analysis system and method based on deep learning |
| CN114366165A (en) * | 2022-01-25 | 2022-04-19 | 上海杏脉信息科技有限公司 | Heart ultrasonic scanning auxiliary system, medium and electronic equipment |
| CN114366165B (en) * | 2022-01-25 | 2024-03-26 | 上海杏脉信息科技有限公司 | Heart ultrasonic scanning auxiliary system, medium and electronic equipment |
| CN115760851A (en) * | 2023-01-06 | 2023-03-07 | 首都儿科研究所附属儿童医院 | Ultrasonic image data processing method and system based on machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109925002A (en) | Artificial intelligence echocardiogram data collection system and its collecting method | |
| US11200693B2 (en) | Image processing | |
| Gong et al. | Fetal congenital heart disease echocardiogram screening based on DGACNN: adversarial one-class classification combined with video transfer learning | |
| CN107730497A (en) | A kind of plaque within blood vessels property analysis method based on depth migration study | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN108830826A (en) | A kind of system and method detecting Lung neoplasm | |
| CN110197713A (en) | A kind of processing method of medical imaging, device, equipment and medium | |
| CN109192305B (en) | Heart function automatic analysis method based on deep circulation neural network | |
| CN112446862B (en) | Dynamic breast ultrasound video full-focus real-time detection and segmentation device and system based on artificial intelligence and image processing method | |
| CN109512464A (en) | A kind of disorder in screening and diagnostic system | |
| CN109620293B (en) | Image recognition method and device and storage medium | |
| CN110363772B (en) | Heart MRI segmentation method and system based on antagonistic network | |
| Samiappan et al. | Classification of carotid artery abnormalities in ultrasound images using an artificial neural classifier. | |
| CN110604597B (en) | Method for intelligently acquiring fetal cardiac cycle images based on ultrasonic four-cavity cardiac section | |
| CN111275706A (en) | Shear wave elastic imaging-based ultrasound omics depth analysis method and system | |
| CN111297399A (en) | Fetal heart positioning and fetal heart rate extraction method based on ultrasonic video | |
| Lin et al. | Echocardiography-based AI detection of regional wall motion abnormalities and quantification of cardiac function in myocardial infarction | |
| CN108877927A (en) | A kind of medical image diagnosis method | |
| CN114581425B (en) | Myocardial segment defect image processing method based on deep neural network | |
| Zamzmi et al. | Echo doppler flow classification and goodness assessment with convolutional neural networks | |
| Mehta et al. | Heart Disease Diagnosis using Deep Learning | |
| CN117036302B (en) | Method and system for determining calcification degree of aortic valve | |
| CN114271850B (en) | Ultrasonic detection data processing method and ultrasonic detection data processing device | |
| Tian et al. | A Heart Segmentation Algorithm Based on Dynamic Ultrasound | |
| Chandra et al. | Mitral Valve Abnormality Detection by fully end to end with Deep Neural Network |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190625 |