CN106408566B - A kind of fetal ultrasound image quality control method and system - Google Patents
A kind of fetal ultrasound image quality control method and system Download PDFInfo
- Publication number
- CN106408566B CN106408566B CN201610991842.7A CN201610991842A CN106408566B CN 106408566 B CN106408566 B CN 106408566B CN 201610991842 A CN201610991842 A CN 201610991842A CN 106408566 B CN106408566 B CN 106408566B
- Authority
- CN
- China
- Prior art keywords
- roi
- index
- window
- image
- meets
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- 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
- G06T2207/30044—Fetus; Embryo
-
- 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
- G06T2207/30168—Image quality inspection
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)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
Fetal ultrasound image quality control method provided by the invention and system, the Fetus Abdominal scanning section gray level image obtained by input clinical ultrasound doctor, it extracts its local phase feature and obtains corresponding Local Symmetric phase and asymmetric phase diagram, RGB image is combined into together with former gray level image, utilize L-CNN model orientation to ROI, and judge whether the ROI in section meets the 1st index, then C-CNN model analyzes the ROI recognized, and judge whether SB meets the 2nd index, whether UV meets the 3rd standard, to realize the control of fetal ultrasound picture quality, fetal ultrasound image quality control method provided by the present application and system, it is controlled by the quantification quality to fetal ultrasound scanning section Automation grade point, the purpose of objective evaluation section quality may be implemented , more more economical than expert's manual operation effective.
Description
Technical field
The present invention relates to medical image technical fields, more particularly, to a kind of fetal ultrasound image quality control method and are
System.
Background technique
The advantages that ultrasound is because of acquisition, at low cost, radiationless and Mobile portable in real time, is widely used in antenatal exaination.It is logical
Often, the process of Prenatal ultrasonography for diagnosing includes: ultrasonic probe scanning, standard section obtains, biometric measures, gestational age weight is surveyed
It is fixed.Biometric measurement is necessary to gestational age body weight determination, and Seeking Truth carries out on standard section again for biometric measurement,
This just illustrates that only the measured value of standard section can accurately just estimate gestational age and weight, it is seen that the importance of standard section.
But not all doctor can obtain standard section.If Fig. 1 (a) is that the fetal abdominal circumference met the quality standard is cut
Face, the length that grey coil measures abdominal circumference (AC) is 338.72mm;The fetus of same gestational age, it is not abundant enough to might have experience
Doctor obtains the fetal abdominal circumference section that quality standard is not met shown in Fig. 1 (b), and the length that grey coil measures AC is
326.87mm;Two images only difference is that arrow meaning magenblase (SB), the full display of (a) magenblase and interior echoless,
(b) there are 11.85mm for the abdominal circumference length that the not good enough display of magenblase, the abdominal circumference length for causing not good enough section to measure and standard section measure
Error.
Existing research highlights the importance of the quality control of fetus biometric measurement in clinic.Generally by experienced
Doctor (expert) scores to ultrasonic scan section according to agreement, realizes the purpose of quality control, but scored manually by expert
Quality control has disadvantages that: be 1. more likely to the subjectivity of expert 2. heavy dependence expert experience 3. repetitive operation work,
It takes time and effort, can not be realized in clinic.
Summary of the invention
Have in view of that, it is necessary to for defect existing for present technology, provide a kind of fetal ultrasound picture quality controlling party
Method and system.
To achieve the above object, the present invention adopts the following technical solutions:
A kind of fetal ultrasound image quality control method, includes the following steps:
Step S110: the gray level image of Fetus Abdominal scanning section is obtained;
Step S120: extracting the local phase feature of the gray level image, obtains corresponding Local Symmetric phase diagram drawn game
The asymmetric phase diagram in portion;
Step S130: the symmetrical phase figure and asymmetric phase diagram and the gray level image are combined into RGB image;
Step S140: using L-CNN model orientation to the ROI in the RGB image, and judge whether ROI meets the 1st
Index;
Step S150: analyzing the ROI recognized using C-CNN model, and judges whether SB meets the 2nd and refer to
Whether mark, UV meet the 3rd standard;
Step S160: fetal ultrasound image quality measurement knot is obtained according to the judging result in step S140 and step S150
Fruit;
Wherein, the 1st index is that the area of ROI accounts for 1/2 or more of scanning sector, and the 2nd index is filled for SB
It is full of display, sharpness of border, interior echoless, the 3rd index is that UV is echo in hook solid shows, do not interrupt continuously, allowing.
In some embodiments, in step S140, using L-CNN model orientation to the ROI in the RGB image, and sentence
Whether disconnected ROI meets the 1st index, specifically include the following steps:
Step S141: the subgraph of each fetal abdominal circumference section is taken off;
Step S142: adjusting the size of the subgraph to 227*227, and input L-CNN, and obtaining window is the general of ROI
Rate;
Step S143: the above-mentioned window probability obtained in whole section centered on each pixel according to this is used, ROI is obtained
Corresponding probability graph;
Step S144: the probability graph is smooth by two-sided filter, probability is obtained most using non-maxima suppression
High pixel is the best center point coordinate C that ROI is navigated to fixed size windowx,y;
Step S145: the ratio R that ROI area accounts for ultrasonic scan sector area is calculatedr/FOVObtain SROI, judge whether ROI accords with
Close the 1st index.
In some embodiments, in step S141, the subgraph of each fetal abdominal circumference section is taken off, using fixed dimension
The window of 320*280 takes off subgraph in ultrasonic scan region with the sliding of stride 10.
In some embodiments, the step S144 further includes following step:
With best center point coordinate Cx,yCentered on, and its be with stride in the regional scope of 10 pixels up and down
10 sliding windows obtain central point, width, the height of highest ROI probability, complete the automatic positioning of ROI in current test image,
In, for the width of the window in the range of 210-400, window height is in the range of 190-340.
In some embodiments, wherein in step S160, obtained according to the judging result in step S140 and step S150
Fetal ultrasound picture quality control result, specifically include the following steps:
If ROI meets the 1st index, to the scoring S of ROIROIIt is 1, if it is not, being denoted as 0;
If SB meets the 2nd index, to the scoring S of SBSBIt is 1, if it is not, being denoted as 0;
If UV meets the 3rd standard, to the scoring S of UVUVIt is 1, if it is not, being denoted as 0;
The then assessment score S of fetal ultrasound picture qualityFAPFor SFAP=SROI+SSB+SUV。
In some embodiments, wherein the width of the window is interval variation, window with 10 in the range of 210-400
Open height is interval variation with 10 in the range of 190-340.
In addition, present invention also provides a kind of fetal ultrasound picture quality controling systems, comprising:
Image collection module, for obtaining the gray level image of Fetus Abdominal scanning section;
Characteristic extracting module obtains corresponding Local Symmetric phase for extracting the local phase feature of the gray level image
Bitmap and local asymmetry phase diagram;
Image co-registration module, for the symmetrical phase figure and asymmetric phase diagram to be combined into the gray level image
RGB image;
L-CNN model orientation module for utilizing the ROI in RGB image described in L-CNN model orientation, and judges that ROI is
It is no to meet the 1st index;
C-CNN model analysis module analyzes the ROI recognized using C-CNN model, and judges whether SB meets
Whether the 2nd index, UV meet the 3rd standard;And
As a result output module, for according to the judgement in the L-CNN model orientation module and C-CNN model analysis module
As a result fetai ultrasonogram image quality measurements are obtained;
Wherein, the 1st index is that the area of ROI accounts for 1/2 or more of scanning sector, and the 2nd index is filled for SB
It is full of display, sharpness of border, interior echoless, the 3rd index is that UV is echo in hook solid shows, do not interrupt continuously, allowing.
In some embodiments, the L-CNN model orientation module includes:
Subgraph takes off unit, for taking off the subgraph of each fetal abdominal circumference section;
Image control unit for adjusting the size of the subgraph to 227*227, and inputs L-CNN, and obtaining window is
The probability of ROI;
Window probability acquiring unit, for using the above-mentioned window obtained in whole section centered on each pixel according to this
Probability obtains the corresponding probability graph of ROI;
Center point coordinate positioning unit, for the probability graph is smooth by two-sided filter, using non-maximum
Inhibit to obtain the highest pixel of probability to be the best center point coordinate C for navigating to ROI with fixed size windowx,y;And
Judging unit accounts for the ratio R of ultrasonic scan sector area for calculating ROI arear/FOVObtain SROI, judge that ROI is
It is no to meet the 1st index.
In some embodiments, the center point coordinate positioning unit is also used to best center point coordinate Cx,yCentered on,
And its up and down in the regional scope of 10 pixels, with stride for 10 sliding windows, the center of highest ROI probability is obtained
Point, width, height complete the automatic positioning of ROI in current test image, wherein range of the width of the window in 210-400
Interior, window height is in the range of 190-340.
In some embodiments, the result output module is used for according to the L-CNN model orientation module and C-CNN mould
Judging result in type analysis module obtains fetai ultrasonogram image quality measurements, comprising: if ROI meets the 1st index,
To the scoring S of ROIROIIt is 1, if it is not, being denoted as 0;
If SB meets the 2nd index, to the scoring S of SBSBIt is 1, if it is not, being denoted as 0;
If UV meets the 3rd standard, to the scoring S of UVUVIt is 1, if it is not, being denoted as 0;
The then assessment score S of fetal ultrasound picture qualityFAPFor SFAP=SROI+SSB+SUV。
The present invention by adopting the above technical scheme the advantages of be:
Fetal ultrasound image quality control method provided by the invention and system are obtained by input clinical ultrasound doctor
Fetus Abdominal scanning section gray level image extracts its local phase feature and obtains corresponding Local Symmetric phase and asymmetric phase
Figure, is combined into RGB image together with former gray level image, is navigated to ROI in the way of multi-resolution scanning by L-CNN model,
And judge whether the ROI in section meets the 1st index, then C-CNN model analyzes the ROI recognized, and judges
Whether SB meets the 2nd index, and whether UV meets the 3rd standard, wherein the area that the 1st index is ROI accounts for scanning fan
1/2 or more of area, the 2nd index are that SB fills display, sharpness of border, interior echoless, and it is curved that the 3rd index, which is UV,
It is hook-shaped to show, do not interrupt continuously, allowing interior echo, to realize the control of fetal ultrasound picture quality, tire provided by the present application
Youngster's ultrasonograph quality control method and system pass through the quantification quality control to fetal ultrasound scanning section Automation grade point
System, may be implemented the purpose of objective evaluation section quality, more more economical than expert's manual operation, efficient;Secondly, can be used in
It is easy to use in tele-medicine;Further more, can be used for the training and study of new hand doctor, by assessing the section of their acquisitions,
The ability that they obtain standard section is continuously improved.
Detailed description of the invention
What (a) was indicated in Fig. 1 is the fetal abdominal circumference section met the quality standard;
What (b) was indicated in Fig. 1 is the fetal abdominal circumference section for not meeting quality standard;
Fig. 2 is the step flow chart of fetal ultrasound image quality control method provided in an embodiment of the present invention;
(a) Fetus Abdominal standard section in Fig. 3, (b) and (c) indicates bright symmetrical and asymmetric phase property, (d) with
(e) dark symmetrical and asymmetric phase property is indicated;
Fig. 4 provides for the application using the ROI in RGB image described in L-CNN model orientation, and judges whether ROI meets
The step flow chart of 1st index;
Fig. 5 is expressed as four kinds of classifications of ROI;
Fig. 6 is expressed as the structural schematic diagram of fetal ultrasound picture quality controling system provided by the present application;
Fig. 7 indicates the structural schematic diagram of the L-CNN model orientation module;
Fig. 8 is FUIQA system provided by the present application and expert E1, E2, and the consistent Fetus Abdominal ultrasound of E3 quality evaluation is swept
Look into section display diagram.
Fig. 9 illustrates FUIQA system and expert E1, E2, the inconsistent Fetus Abdominal ultrasonic scan section of E3 assessment result
Schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing and specific implementation
Example, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only to explain this hair
It is bright, it is not intended to limit the present invention.
In application documents, relational terms such as first and second and the like are used merely to an entity or operation
It is distinguished with another entity or operation, without necessarily requiring or implying between these entities or operation, there are any this
Actual relationship or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to nonexcludability
It include so that the process, method, article or equipment for including a series of elements not only includes those elements, but also to wrap
Include other elements that are not explicitly listed, or further include for this process, method, article or equipment intrinsic want
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described want
There is also other identical elements in the process, method, article or equipment of element.
Referring to Fig. 2, a kind of fetal ultrasound image quality control method provided in an embodiment of the present invention, including following steps
It is rapid:
Step S110: the gray level image of Fetus Abdominal scanning section is obtained;
In the present embodiment, input picture provided by the present application is the Fetus Abdominal scanning section that clinical ultrasound doctor obtains
Gray level image, it will be understood that in practical applications, can be extended to other ultrasonic scan sections of fetus.
Step S120: extracting the local phase feature of the gray level image, obtains corresponding Local Symmetric phase diagram drawn game
The asymmetric phase diagram in portion;
Referring to Fig. 3, the bright and dark phase for illustrating the extraction of a Fetus Abdominal standard section is symmetrical with asymmetric spy
Sign figure, wherein (a) Fetus Abdominal standard section in Fig. 3, (b) and (c) indicates bright symmetrical and asymmetric phase property, (d) with
(e) dark symmetrical and asymmetric phase property is indicated.
Step S130: the symmetrical phase figure and asymmetric phase diagram and the gray level image are combined into RGB image;
It is appreciated that natural image is the color image in tri- channels RGB, and ultrasound image is single pass grayscale image
Picture, in order to analyze ultrasound image using DCNN model, it will usually which replicating single-channel data to increase is 3 channels, and Local Phase
Position feature is effectively applied to many ultrasound images processing problems, and in order to increase gray level image information, we are by Local Symmetric
Optional channel with asymmetric phase property as original image inputs, and is combined into RGB image with the gray level image.
Step S140: using the ROI (abdomen area) in RGB image described in L-CNN model orientation, and whether judge ROI
Meet the 1st index;Wherein, the 1st index is that the area of ROI accounts for 1/2 or more of scanning sector;
Table 1 is please referred to, for the fetal abdominal circumference section quality control protocol executed in the application.
Index | Anatomical structure | Standard |
1st index | Abdomen area (ROI) | Area accounts for 1/2 or more of scanning sector |
2nd index | Magenblase (SB) | Full display, sharpness of border, interior echoless |
3rd index | Umbilical vein (UV) | Hook solid shows, does not interrupt continuously, allowing interior echo |
Depth convolutional network (DCNN) model is a kind of depth learning technology of hot topic, has been successfully applied to medical image
The various problems of analysis.Typical DCNN structure is by several groups of convolutional layers, maximum value pond layer and the last full articulamentum group closely followed
At.Convolutional layer extracts feature using upper one layer of local receptor field as input, and retains global spatial structural form, this Shen
The L-CNN that please be proposed is based on the basis of DCNN.
Table 1 is please referred to, is L-CNN model structure.In the present embodiment, for preferably transfer learning natural image data
Trained parameter, we by the convolutional layer of L-CNN network and the structure setting of pond layer at as AlexNet model, and
Subsequent 3 full articulamentums are finely tuned on ultrasonic training data, it is single that this 3 full articulamentums separately include 1024,256,2 nerves
Member, that is to say, that during L-CNN training, the parameter of convolutional layer is as at the beginning of trained AlexNet model parameter
Beginningization, and then random initializtion is Gaussian Profile to the parameter of full articulamentum.Meanwhile in order to improve generalization ability, dropout plan
Slightly and ReLu is used for training process.Learning rate is initially 0.001, and is gradually reduced 10 times when network stops and restraining.
The structure of 1 L-CNN model of table
Referring to Fig. 4, it is provided by the present application, in step S140, using in RGB image described in L-CNN model orientation
ROI, and judge whether ROI meets the 1st index, specifically include the following steps:
Step S141: the subgraph of each fetal abdominal circumference section is taken off;
Preferably for each fetal abdominal circumference section, in the region of ultrasonic scan, with fixed dimension 320*280's
Window takes off subgraph with the sliding of stride 10.It is appreciated that can also be using other windows for different fetal abdominal circumference sections
Mouth and stride take off subgraph.
Step S142: adjusting the size of the subgraph to 227*227, and input L-CNN, and obtaining window is the general of ROI
Rate;
Step S143: the above-mentioned window probability obtained in whole section centered on each pixel according to this is used, ROI is obtained
Corresponding probability graph;
Step S144: the probability graph is smooth by two-sided filter, probability is obtained most using non-maxima suppression
High pixel is the best center point coordinate C that ROI is navigated to fixed size windowx,y;
Further, the step S144 further includes following step:
With best center point coordinate Cx,yCentered on, and its be with stride in the regional scope of 10 pixels up and down
10 sliding windows obtain central point, width, the height of highest ROI probability, complete the automatic positioning of ROI in current test image,
In, for the width of the window in the range of 210-400, window height is in the range of 190-340.
Step S145: the ratio R that ROI area accounts for ultrasonic scan sector area is calculatedr/FOVObtain SROI, judge whether ROI accords with
Close the 1st index.
It is appreciated that S141~step S145 can complete to scheme using RGB described in L-CNN model orientation through the above steps
ROI as in, and judge whether ROI meets the 1st index.
Step S150: analyzing the ROI recognized using C-CNN model, and judges whether SB meets the 2nd and refer to
Whether mark, UV meet the 3rd standard, and the 2nd index is that SB fills display, sharpness of border, interior echoless, and described 3rd
Index is that UV is echo in hook solid shows, do not interrupt continuously, allowing.
It is appreciated that the purpose of C-CNN model is to realize whether two important anatomy structures of SB and UV meet matter inside ROI
The assessment of amount standard.Clinically, be anatomical structure is divided into standard, it is not good enough, there is no three kinds of judge forms, it is not good enough to represent the solution
Cut open structure exist but it is nonstandard.In fact, the section of only standard state is just recognized for biometric measurement, it is not good enough
There are conspicuousness errors in biometric measurement with standard section for the section of state.
It is appreciated that in this application, only anatomical structure standard when, the scoring of anatomical structure is just 1 in corresponding table 1;
It is not good enough be not present in the case where, the scoring of the anatomical structure is 0.Meanwhile whether we meet the feelings of standard according to SB and UV
All ROI are divided into 4 kinds of classifications by condition, and the quality evaluation problem of C-CNN is transformed into 4 classification problems, as shown in figure 5, indicating
For four kinds of classifications of ROI.
It is assessed it is appreciated that the 2nd and the 3rd in table 1 is integrated and is transformed into the modes of 4 classification, is to utilize C-
CNN learns the feature of abdomen area entirety, and is more than and learns single anatomical structure.Simultaneously as three indexs are phases in table 1
Mutual correlation, L-CNN model parameter transfer learning is assisted the classification of 4 classifications by we to C-CNN model.The instruction of C-CNN
Practice details and structure is similar to L-CNN, in addition to the neural unit number of last 3 full articulamentums is respectively 2048,256,4.C-
After CNN model training is completed, it is input to C-CNN after the ROI being automatically positioned of L-CNN in test image is taken off, obtains Fig. 5
The shown other probability of 4 type, maximum probability is classification belonging to the ROI, completes the 2nd and the 3rd index in table 1 oneself
Dynamic assessment.
Step S160: fetal ultrasound image quality measurement knot is obtained according to the judging result in step S140 and step S150
Fruit;
It is appreciated that according to table 1, the 1st is the area ratio for assessing abdomen area, thus by L-CNN precise positioning to
After ROI, the ratio R that ROI area accounts for ultrasonic scan sector area is calculatedr/FOVObtain SROI, see formula (1).It is later that ROI is defeated
Enter C-CNN, obtain the classification of ROI, according to the corresponding scoring for respectively obtaining UV and SB of Fig. 5, i.e. SSBAnd SUV.And then obtain one
The quality evaluation score S of ultrasonic scan sectionFAP, see formula (2).
SFAP=SROI+SSB+SUV (2)
Although frame has general it is appreciated that we are currently only used for the quality of assessment Fetus Abdominal scanning section
Property, the quality evaluation of other sections of fetal ultrasound scanning can also be extended to.
Referring to Fig. 6, fetal ultrasound picture quality controling system provided by the present application, comprising: image collection module 110,
For obtaining the gray level image of Fetus Abdominal scanning section;
Characteristic extracting module 120 obtains corresponding Local Symmetric for extracting the local phase feature of the gray level image
Phase diagram and local asymmetry phase diagram;Image co-registration module 130, for by the symmetrical phase figure and asymmetric phase diagram with
The gray level image is combined into RGB image;L-CNN model orientation module 140, for being schemed using RGB described in L-CNN model orientation
ROI as in, and judge whether ROI meets the 1st index;C-CNN model analysis module 150, using C-CNN model to identification
To ROI analyzed, and judge whether SB meets the 2nd index, whether UV meets the 3rd standard;And result output module
160, for obtaining fetal ultrasound according to the judging result in the L-CNN model orientation module and C-CNN model analysis module
Image quality measurement result;Wherein, the 1st index is that the area of ROI accounts for 1/2 or more of scanning sector, and described 2nd refers to
It is designated as SB and fills display, sharpness of border, interior echoless, the 3rd index is that UV is that hook solid shows, continuously do not interrupt, permits
Perhaps interior echo.
Wherein, referring to Fig. 7, the L-CNN model orientation module 140 includes: that subgraph takes off unit 141, for detaining
Take the subgraph of each fetal abdominal circumference section;Image control unit 142 is used to adjust the size of the subgraph to 227*
227, and L-CNN is inputted, obtain the probability that window is ROI;Window probability acquiring unit 143 be used for using it is above-mentioned obtain according to this it is whole
The window probability in section centered on each pixel is opened, the corresponding probability graph of ROI is obtained;Center point coordinate positioning unit 144
For the probability graph is smooth by two-sided filter, obtaining the highest pixel of probability using non-maxima suppression is
The best center point coordinate C of ROI is navigated to fixed size windowx,y;And judging unit 145 accounts for ultrasound for calculating ROI area
The ratio R of scanning sector arear/FOVObtain SROI, judge whether ROI meets the 1st index.
Preferably, the center point coordinate positioning unit 144 is also used to best center point coordinate Cx,yCentered on, and its
Up and down in the regional scope of 10 pixels, with stride for 10 sliding windows, obtain the central point of highest ROI probability, width,
Height completes the automatic positioning of ROI in current test image, wherein the width of the window is in the range of 210-400, window
Open height is in the range of 190-340.
Preferably, the result output module 160 is used for according to the L-CNN model orientation module and C-CNN model point
Analysis module in judging result obtain fetai ultrasonogram image quality measurements, if including ROI meet the 1st index, to ROI
Scoring SROIIt is 1, if it is not, being denoted as 0;
If SB meets the 2nd index, to the scoring S of SBSBIt is 1, if it is not, being denoted as 0;
If UV meets the 3rd standard, to the scoring S of UVUVIt is 1, if it is not, being denoted as 0;
The then assessment score S of fetal ultrasound picture qualityFAPFor SFAP=SROI+SSB+SUV。
Fetal ultrasound image quality control method provided by the invention and system are obtained by input clinical ultrasound doctor
Fetus Abdominal scanning section gray level image extracts its local phase feature and obtains corresponding Local Symmetric phase and asymmetric phase
Figure, is combined into RGB image together with former gray level image, is navigated to ROI in the way of multi-resolution scanning by L-CNN model,
And judge whether the ROI in section meets the 1st index, then C-CNN model analyzes the ROI recognized, and judges
Whether SB meets the 2nd index, and whether UV meets the 3rd standard, wherein the area that the 1st index is ROI accounts for scanning fan
1/2 or more of area, the 2nd index are that SB fills display, sharpness of border, interior echoless, and it is curved that the 3rd index, which is UV,
It is hook-shaped to show, do not interrupt continuously, allowing interior echo, to realize the control of fetal ultrasound picture quality, tire provided by the present application
Youngster's ultrasonograph quality control method and system pass through the quantification quality control to fetal ultrasound scanning section Automation grade point
System, may be implemented the purpose of objective evaluation section quality, more more economical than expert's manual operation, efficient;Secondly, can be used in
It is easy to use in tele-medicine;Further more, can be used for the training and study of new hand doctor, by assessing the section of their acquisitions,
The ability that they obtain standard section is continuously improved.
Technical solution provided by the invention is described in detail below in conjunction with specific embodiment.
We study all Fetus Abdominal ultrasonic scans section used and both are from Shenzhen healthcare hospital for women & children, acquisition time
From in September, 2012 in November, 2013.According to the antenatal exaination agreement of standard, ultrasound cross-section ties up ultrasonic spy by traditional portable 2
Head obtains pregnant woman's scanning in 16-40 weeks pregnant age.The instrument of all acquiring ultrasound images is Siemens Acuson Sequoia
512 ultrasound scanners, but include different parameter settings.In our study, 492 sets of ultrasonic scans of acquisition in 2012
Video is used as training dataset, and 219 sets of ultrasonic scan videos of acquisition in 2013 are used as test data set.
We concentrate the training for being extracted 8072 scanning sections for L-CNN model from 492 sets of training datas, and each
Section take off its abdomen area generate a ROI positive sample, by section with the Duplication of positive sample in 40% work below
For negative sample, negative sample is taken off immediately in background area.Training of the positive sample of L-CNN model training as C-CNN model
Sample, corresponding classification 1,2,3,4 have 1256,3827,151,2838 samples respectively, for the balance of 4 class sample datas, we
Angular range inward turning by the sample of classification 1,3,4 in (- 25,25), which transfers, increases its sample size.Therefore, final C-CNN model
4 trained classification sample sizes are respectively 3717,3827,3907,3964, and this 15415 samples are all used as L- in total
The positive sample of CNN model training.All training datas are marked has the Ultrasonography postgraduate of 2 years rich experiences complete by one
At, while by the examination of our clinical guidance committee and revising the accuracy to ensure data.In training L-CNN and C-CNN
During model, training sample is also divided into training set according to the ratio of 9:1 for we and verifying collects, and detailed details is shown in Table
3。
Table 3.L-CNN and C-CNN training sample number is shown
Ultrasonic scan section for control (FUIQA) system testing of fetal ultrasound picture quality has 2606, comes from 66
219 sets of ultrasonic scan videos of a pregnant woman's acquisition.According to the criterion in table 1, each ultrasound cross-section is all by 3 working experiences 3
Year or more Ultrasonography attending physician mark the position of ROI region, and judged to whether UV and SB complies with standard.This Shen
Please in use E1 respectively, E2, E3 represent 3 as doctor, and 3 doctors are complete independently labels.
By FUIQA system, in ROI, SB and UV, individually the label of each assessment result respectively with 3 experts is opposed for we
Than taking accuracy rate (Acc), sensibility (Sen) and specificity (Spec) quantitatively to show the performance of L-CNN and C-CNN, such as table respectively
Shown in 4.
Table 4. is directed to single anatomical structure ROI, SB, UV, and quantitative pair of FUIQA system and 3 expert E1, E2, E3 scoring
Than
Find out from above-mentioned experimental section, the performance of the FUIQA system of proposition provided by the present application, Fig. 8, which is also shown, to be based on
The FUIQA system of DCNN different parameters is arranged the robustness of image, meanwhile, by (b) in Fig. 8, (d) can be seen that with (h)
FUIQA system also can correctly control picture quality in the case where ultrasound image is seriously affected by sound shadow.Diversified trained number
According to study, the FUIQA system based on DCNN can be considered different images parameter setting, the variation of fetus position, sound shadow and
A variety of situations such as the influence of noise, relatively traditional Image Processing and Pattern Recognition method occupy very big advantage.
Fig. 9 illustrates FUIQA system and expert E1, E2, the inconsistent Fetus Abdominal ultrasonic scan section of E3 assessment result.
Wherein, the position ROI that the automatic positioning of our FUIQA systems and expert mark in (a) figure in Fig. 9 is very close, but obtains not
Consistent score.The R of E1, E2, E3 and FUIQA systemr/FOVRespectively 0.497,0.513,0.476 and 0.523, even if four is non-
Very close to, but ROI scores are calculated according to formula (4).(b) by sound shadow, 3 experts still may recognize that the interlude of UV in figure
The crotch-shaped of UV, and our FUIQA systems judge that UV is nonstandard because that can not identify UV interlude, obtain with 3 experts not
Consistent scoring.
Certainly fetal ultrasound image quality control method and method of the invention can also have a variety of transformation and remodeling, not
It is confined to the specific structure of above embodiment.In short, protection scope of the present invention should include those skills common for this field
It obviously converts or substitutes and retrofit for art personnel.
Claims (10)
1. a kind of fetal ultrasound image quality control method, which is characterized in that include the following steps:
Step S110: the gray level image of Fetus Abdominal scanning section is obtained;
Step S120: extracting the local phase feature of the gray level image, obtains corresponding Local Symmetric phase diagram and part is non-
Symmetrical phase figure;
Step S130: the symmetrical phase figure and asymmetric phase diagram and the gray level image are combined into RGB image;
Step S140: using the ROI in RGB image described in L-CNN model orientation, and judge whether ROI meets the 1st index;
Step S150: analyzing the ROI recognized using C-CNN model, and judges whether SB meets the 2nd index, UV
Whether 3rd standard is met;
Step S160: fetai ultrasonogram image quality measurements are obtained according to the judging result in step S140 and step S150;
Wherein, the area that the 1st index is ROI accounts for 1/2 or more of scanning sector, and the 2nd index is SB full aobvious
Show, sharpness of border, interior echoless, the 3rd index is that UV is echo in hook solid shows, do not interrupt continuously, allowing;
The L-CNN is convolutional neural networks used for positioning, and the C-CNN is the convolutional neural networks for classification, described
RO I is abdomen area, and the SB is magenblase, and the UV is umbilical vein.
2. fetal ultrasound image quality control method according to claim 1, which is characterized in that in step S140, utilize
ROI in RGB image described in L-CNN model orientation, and judge whether ROI meets the 1st index, specifically include the following steps:
Step S141: the subgraph of each fetal abdominal circumference section is taken off;
Step S142: adjusting the size of the subgraph to 227*227, and input L-CNN, obtains the probability that window is ROI;
Step S143: the window probability in whole section centered on each pixel is obtained, the corresponding probability graph of ROI is obtained;
Step S144: the probability graph is smooth by two-sided filter, it is highest that probability is obtained using non-maxima suppression
The coordinate of pixel is the best center point coordinate C that ROI is navigated to fixed size windowx,y;
Step S145: the ratio R that ROI area accounts for ultrasonic scan sector area is calculatedr/FOVObtain the scoring S to ROIROI, judgement
Whether ROI meets the 1st index.
3. fetal ultrasound image quality control method according to claim 2, which is characterized in that in step S141, take off
The subgraph of each fetal abdominal circumference section uses the window of fixed dimension 320*280 with the sliding of stride 10, in ultrasonic scan area
Subgraph is taken off in domain.
4. fetal ultrasound image quality control method according to claim 2, which is characterized in that the step S144 is also wrapped
Include following step:
With best center point coordinate Cx,yCentered on, and its be 10 sliding with stride up and down in the regional scope of 10 pixels
Dynamic window, obtains central point, width, the height of the window of highest ROI probability, completes the automatic positioning of ROI in current test image,
Wherein, the width of the window is obtained in the range of 210-400, and window height is in the range of 190-340.
5. fetal ultrasound image quality control method according to claim 2, which is characterized in that wherein, in step S160,
Fetai ultrasonogram image quality measurements are obtained according to the judging result in step S140 and step S150, specifically include following steps
It is rapid:
If ROI meets the 1st index, to the scoring S of ROIROIIt is 1, if it is not, being denoted as 0;
If SB meets the 2nd index, to the scoring S of SBSBIt is 1, if it is not, being denoted as 0;
If UV meets the 3rd standard, to the scoring S of UVUVIt is 1, if it is not, being denoted as 0;
The then assessment score S of fetal ultrasound picture qualityFAPFor SFAP=SROI+SSB+SUV。
6. fetal ultrasound image quality control method according to claim 4, which is characterized in that wherein, the window
Width is interval variation with 10 in the range of 210-400, and window height is interval variation with 10 in the range of 190-340.
7. a kind of fetal ultrasound picture quality controling system characterized by comprising
Image collection module, for obtaining the gray level image of Fetus Abdominal scanning section;
Characteristic extracting module obtains corresponding Local Symmetric phase diagram for extracting the local phase feature of the gray level image
With local asymmetry phase diagram;
Image co-registration module, for the symmetrical phase figure and asymmetric phase diagram and the gray level image to be combined into RGB figure
Picture;
Whether L-CNN model orientation module for utilizing L-CNN model orientation to the ROI in the RGB image, and judges ROI
Meet the 1st index;
C-CNN model analysis module analyzes the ROI recognized using C-CNN model, and judges whether SB meets the 2nd
Whether item index, UV meet the 3rd standard;And
As a result output module, for according to the judging result in the L-CNN model orientation module and C-CNN model analysis module
Obtain fetai ultrasonogram image quality measurements;
Wherein, the area that the 1st index is ROI accounts for 1/2 or more of scanning sector, and the 2nd index is SB full aobvious
Show, sharpness of border, interior echoless, the 3rd index is that UV is echo in hook solid shows, do not interrupt continuously, allowing;
The L-CNN convolutional neural networks used for positioning, convolutional neural networks of the C-CNN for classification, the ROI are
Abdomen area, the SB are magenblase, and the UV is umbilical vein.
8. fetal ultrasound picture quality controling system according to claim 7, which is characterized in that the L-CNN model is fixed
Position module include:
Subgraph takes off unit, for taking off the subgraph of each fetal abdominal circumference section;
Image control unit for adjusting the size of the subgraph to 227*227, and inputs L-CNN, and obtaining window is ROI
Probability;
Window probability acquiring unit, for using, above-mentioned subgraph takes off unit and above-mentioned image control unit obtains whole section
Window probability centered on interior each pixel obtains the corresponding probability graph of ROI;
Center point coordinate positioning unit, for the probability graph is smooth by two-sided filter, using non-maxima suppression
The coordinate for obtaining the highest pixel of probability is the best center point coordinate C that ROI is navigated to fixed size windowx,y;And
Judging unit accounts for the ratio R of ultrasonic scan sector area for calculating ROI arear/FOVObtain the scoring S to ROIROI, sentence
Whether disconnected ROI meets the 1st index.
9. fetal ultrasound picture quality controling system according to claim 8, which is characterized in that the center point coordinate is fixed
Bit location is also used to best center point coordinate Cx,yCentered on, and its up and down in the regional scope of 10 pixels, with step
Width is 10 sliding windows, obtains central point, width, the height of the window of highest ROI probability, completes ROI in current test image
Automatic positioning, wherein for the width of the window in the range of 210-400, window height is in the range of 190-340.
10. fetal ultrasound picture quality controling system according to claim 7, which is characterized in that the result exports mould
Block is used to obtain fetai ultrasonogram according to the judging result in the L-CNN model orientation module and C-CNN model analysis module
Image quality measurements, if including ROI meet the 1st index, to the scoring S of ROIROIIt is 1, if it is not, being denoted as 0;
If SB meets the 2nd index, to the scoring S of SBSBIt is 1, if it is not, being denoted as 0;
If UV meets the 3rd standard, to the scoring S of UVUVIt is 1, if it is not, being denoted as 0;
The then assessment score S of fetal ultrasound picture qualityFAPFor SFAP=SROI+SSB+SUV。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610991842.7A CN106408566B (en) | 2016-11-10 | 2016-11-10 | A kind of fetal ultrasound image quality control method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610991842.7A CN106408566B (en) | 2016-11-10 | 2016-11-10 | A kind of fetal ultrasound image quality control method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106408566A CN106408566A (en) | 2017-02-15 |
CN106408566B true CN106408566B (en) | 2019-09-10 |
Family
ID=59230646
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610991842.7A Active CN106408566B (en) | 2016-11-10 | 2016-11-10 | A kind of fetal ultrasound image quality control method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106408566B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108056789A (en) * | 2017-12-19 | 2018-05-22 | 飞依诺科技(苏州)有限公司 | A kind of method and apparatus for the configuration parameter value for generating ultrasound scanning device |
CN108389174B (en) * | 2018-03-30 | 2021-06-08 | 苏州佳世达电通有限公司 | Ultrasonic imaging system and ultrasonic imaging method |
CN109145830B (en) * | 2018-08-24 | 2020-08-25 | 浙江大学 | Intelligent water gauge identification method |
CN109191442B (en) * | 2018-08-28 | 2021-04-13 | 深圳大学 | Ultrasonic image evaluation and screening method and device |
CN111820948B (en) * | 2019-04-20 | 2022-03-18 | 无锡祥生医疗科技股份有限公司 | Fetal growth parameter measuring method and system and ultrasonic equipment |
CN110111329B (en) * | 2019-05-17 | 2021-05-11 | 四川大学华西第二医院 | Artificial intelligence based ultrasonic image detection method and system |
CN110464380B (en) * | 2019-09-12 | 2021-10-29 | 李肯立 | Method for controlling quality of ultrasonic section image of fetus in middle and late pregnancy |
CN111127305B (en) * | 2019-10-14 | 2021-01-26 | 李胜利 | Method for automatically obtaining standard tangent plane based on three-dimensional volume of fetal craniofacial part in early pregnancy |
CN112634272B (en) * | 2021-03-09 | 2021-06-01 | 四川大学 | Mammary gland ultrasonic imaging quality monitoring system and method based on deep neural network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014089647A1 (en) * | 2012-12-14 | 2014-06-19 | Universidade Estadual Paulista "Júlio De Mesquita Filho" | Method for determining embryo viability and quality |
CN104408717A (en) * | 2014-11-24 | 2015-03-11 | 北京航空航天大学 | Pathological image color quality comprehensive evaluation method based on color separation |
CN103239249B (en) * | 2013-04-19 | 2015-04-22 | 深圳大学 | Measurement method for fetus ultrasound image |
CN105979253A (en) * | 2016-05-06 | 2016-09-28 | 浙江科技学院 | Generalized regression neural network based non-reference stereoscopic image quality evaluation method |
-
2016
- 2016-11-10 CN CN201610991842.7A patent/CN106408566B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014089647A1 (en) * | 2012-12-14 | 2014-06-19 | Universidade Estadual Paulista "Júlio De Mesquita Filho" | Method for determining embryo viability and quality |
CN103239249B (en) * | 2013-04-19 | 2015-04-22 | 深圳大学 | Measurement method for fetus ultrasound image |
CN104408717A (en) * | 2014-11-24 | 2015-03-11 | 北京航空航天大学 | Pathological image color quality comprehensive evaluation method based on color separation |
CN105979253A (en) * | 2016-05-06 | 2016-09-28 | 浙江科技学院 | Generalized regression neural network based non-reference stereoscopic image quality evaluation method |
Also Published As
Publication number | Publication date |
---|---|
CN106408566A (en) | 2017-02-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106408566B (en) | A kind of fetal ultrasound image quality control method and system | |
US20170367685A1 (en) | Method for processing 3d image data and 3d ultrasonic imaging method and system | |
CN111179227B (en) | Mammary gland ultrasonic image quality evaluation method based on auxiliary diagnosis and subjective aesthetics | |
Kim et al. | Machine-learning-based automatic identification of fetal abdominal circumference from ultrasound images | |
US10762630B2 (en) | System and method for structures detection and multi-class image categorization in medical imaging | |
TWI684997B (en) | Establishing method of bone age assessment and height prediction model, bone age assessment and height prediction system, and bone age assessment and height prediction method | |
WO2016194161A1 (en) | Ultrasonic diagnostic apparatus and image processing method | |
CN112465772B (en) | Fundus colour photographic image blood vessel evaluation method, device, computer equipment and medium | |
EP4198819A1 (en) | Method for detecting and classifying lesion area in clinical image | |
Rahmatullah et al. | Quality control of fetal ultrasound images: Detection of abdomen anatomical landmarks using adaboost | |
CN102436551B (en) | Computer aided gastric cancer diagnosis method based on target tracking | |
CN111462049A (en) | Automatic lesion area form labeling method in mammary gland ultrasonic radiography video | |
WO2022267300A1 (en) | Method and system for automatically extracting target area in image, and storage medium | |
CN111950595A (en) | Liver focus image processing method, system, storage medium, program, and terminal | |
CN110543912A (en) | Method for automatically acquiring cardiac cycle video in fetal key section ultrasonic video | |
CN110279433A (en) | A kind of fetus head circumference automatic and accurate measurement method based on convolutional neural networks | |
CN111528907A (en) | Ultrasonic image pneumonia auxiliary diagnosis method and system | |
CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
CN115564741A (en) | Device for quantifying esophageal and gastric fundus varices based on CT (computed tomography) images | |
CN108629770A (en) | Ultrasonic image division method based on support vector machines | |
WO2020215485A1 (en) | Fetal growth parameter measurement method, system, and ultrasound device | |
Xing et al. | Automatic detection of A‐line in lung ultrasound images using deep learning and image processing | |
Zeng et al. | TUSPM-NET: A multi-task model for thyroid ultrasound standard plane recognition and detection of key anatomical structures of the thyroid | |
Li et al. | A supervised method using convolutional neural networks for retinal vessel delineation | |
CN106372593A (en) | Optic disc area position method based on blood vessel convergence |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20210201 Address after: 518000 5/6 Pingshan Private Enterprise Science Park, Taoyuan Street, Nanshan District, Shenzhen City, Guangdong Province Patentee after: SHENZHEN WISONIC MEDICAL TECHNOLOGY Co.,Ltd. Address before: 518000 No. 3688 Nanhai Road, Shenzhen, Guangdong, Nanshan District Patentee before: SHENZHEN University |
|
TR01 | Transfer of patent right |