CN110335267A - A kind of cervical lesions method for detecting area - Google Patents
A kind of cervical lesions method for detecting area Download PDFInfo
- Publication number
- CN110335267A CN110335267A CN201910602980.5A CN201910602980A CN110335267A CN 110335267 A CN110335267 A CN 110335267A CN 201910602980 A CN201910602980 A CN 201910602980A CN 110335267 A CN110335267 A CN 110335267A
- Authority
- CN
- China
- Prior art keywords
- indicate
- candidate region
- region
- image
- characteristic
- 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
- 230000003902 lesion Effects 0.000 title claims abstract description 50
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000010606 normalization Methods 0.000 claims abstract description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 18
- 230000005284 excitation Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 230000006735 deficit Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 208000019065 cervical carcinoma Diseases 0.000 abstract description 11
- 238000001514 detection method Methods 0.000 abstract description 8
- 238000012216 screening Methods 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 4
- QTBSBXVTEAMEQO-UHFFFAOYSA-N Acetic acid Chemical compound CC(O)=O QTBSBXVTEAMEQO-UHFFFAOYSA-N 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 3
- 201000011510 cancer Diseases 0.000 description 3
- 241000701806 Human papillomavirus Species 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 2
- 229910002056 binary alloy Inorganic materials 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 208000005623 Carcinogenesis Diseases 0.000 description 1
- 206010011469 Crying Diseases 0.000 description 1
- 208000018079 Uterine Cervical disease Diseases 0.000 description 1
- 208000006374 Uterine Cervicitis Diseases 0.000 description 1
- 238000001574 biopsy Methods 0.000 description 1
- 210000004204 blood vessel Anatomy 0.000 description 1
- 230000036952 cancer formation Effects 0.000 description 1
- 231100000504 carcinogenesis Toxicity 0.000 description 1
- 206010008323 cervicitis Diseases 0.000 description 1
- 208000026735 cervix disease Diseases 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000000981 epithelium Anatomy 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000009826 neoplastic cell growth Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000003716 rejuvenation Effects 0.000 description 1
- 210000005000 reproductive tract Anatomy 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
-
- 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
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of cervical lesions method for detecting area of cervical lesions detection field, include the following steps: step S10, uterine neck image is shot by gynecatoptron and is sent to computer；Step S20, received image is pre-processed；Step S30, using region candidate network by choosing candidate region on pretreated image；Step S40, recurrence calculating is carried out to candidate region using normalization exponential function, and then cervical lesions region is demarcated.The present invention has the advantages that improving the accuracy in detection of cervical carcinoma lesion region.
Description
Technical field
The present invention relates to cervical lesions detection fields, refer in particular to a kind of cervical lesions method for detecting area.
Background technique
Cervical carcinoma is the fourth-largest malignant tumour of women, is the most common genital tract malignant tumour, the annual cervical carcinoma in the whole world
New cases up to 528,000, wherein dead number of cases is up to 266,000.With the number of the infected and death toll of cervical carcinoma
Increase year by year, and morbidity crowd present rejuvenation so that the prevention and treatment of cervical carcinoma/cervical lesions seems extremely heavy with diagnosis and treatment
It wants.Therefore, carrying out extensive, specification cervical carcinoma screening project for general population is to reduce uterine neck carcinogenesis and death most
One of effective ways.
It carries out diagnosis to cervical carcinoma to need to shoot multiple multi-period uterine neck images using gynecatoptron, gynecatoptron, which has become, to be faced
The important tool of screening CIN (Cervical intraepitheliaI neoplasia) and early cervical carcinoma on bed, and directly affect the diagnosis and treatment scheme of patient.
However, gynecatoptron image is a morphology technology, ununified diagnostic criteria, image specificity is low, even if
Cervicitis or simple HPV (human papilloma virus) infection can also generate abnormal image under gynecatoptron.Abnormal vaginal mirror
The form of expression of image is all kinds of, both can behave as acetic acid white epithelium, can also be special-shaped blood vessel, and various abnormal images can also be same
When occur, therefore the high sensitivity of gynecatoptron diagnostic imaging cervical lesions and specificity is low.And the subjective factor of operator is especially
It is that its level professional technology and clinical experience etc. have large effect to the inspection result of gynecatoptron；The depth of cervical biopsy, model
The same accuracy for influencing diagnosis such as enclose.
Therefore, how a kind of cervical lesions method for detecting area is provided, realizes that the detection for improving cervical carcinoma lesion region is quasi-
Exactness becomes a urgent problem to be solved.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of cervical lesions method for detecting area, realizes and improve uterine neck
The accuracy in detection of cancer lesion region.
The present invention is implemented as follows: a kind of cervical lesions method for detecting area, described method includes following steps:
Step S10, uterine neck image is shot by gynecatoptron and is sent to computer；
Step S20, received image is pre-processed；
Step S30, using region candidate network by choosing candidate region on pretreated image；
Step S40, recurrence calculating carried out to candidate region using normalization exponential function, so to cervical lesions region into
Rower is fixed.
Further, the step S20 is specifically included:
Step S21, received image is normalized；
Step S22, depth characteristic is extracted from the image after normalized.
Further, the step S21 is specifically included:
Step S211, the RGB image for being N*N at size by received image scaling, wherein N > 0；
Step S212, by standard data set Software for producing, cervical lesions region mark is carried out to the RGB image after scaling
Note；
Step S213, the image input convolutional neural networks of mark are created into training pattern.
Further, the step S22 is specifically included:
Step S221, classification foundation of the shallow-layer feature of the training pattern as lesion region is chosen；
Step S222, depth characteristic is extracted to the training pattern:
WhereinIndicate will own in convolutional neural networks
The Feature Mapping that residual error module generates is as input, WLAnd WL-1Successively indicate that two 3 × 3 convolution weight matrixs, BN () indicate
The output data of each hidden layer of convolutional neural networks is normalized, f () indicates ReLU activation primitive,It indicates
The convolution operation of convolutional neural networks；
Step S223, the global draw pond operated by Squeeze, presses the depth characteristic after convolution operation
Contracting, so that the real number ordered series of numbers of C characteristic layer boil down to 1*1*C:
Wherein C indicates that the port number of characteristic layer, W indicate the width of characteristic layer,
H indicates the height of characteristic layer, ucIndicate characteristic layer channel, i, j are positive integer, Fsq() indicates Squeeze operation；
Step S224, it is operated by Excitation by ZcWeight as characteristic layer carries out weight to every layer of depth characteristic
Distribution:
Fex(Zc, W) and=σ (W2δ(W1Zc)), wherein Fex() indicates Excitation operation, and δ indicates ReLU activation primitive, W1
Indicate the parameter that full connection generates for the first time, W2Indicate the parameter that second of full connection generates.
Further, the step S30 is specifically included:
Step S31, by any scale image input area candidate network；
Step S32, described any scale image generates characteristic pattern using the convolution inclusion layer of convolutional neural networks；
Step S33, multiple dimensioned convolution operation is carried out on the characteristic pattern and chooses candidate region, and gives each candidate region
Distribution one for mark whether be lesion region binary label；
Step S34, the set of region candidate network output candidate region.
Further, the step S33 is specifically included:
Step S331, sliding carries out the sliding window of selection feature at random on characteristic pattern for creation one；
Step S332, centered on the center of the sliding window, using 3 kinds of scales and 3 kinds of length-width ratios on characteristic pattern
Map the candidate region of 9 kinds of different scales；
Step S333, to each candidate region distribution one for mark whether be lesion region binary label；
Step S334, judge that the IOU of candidate region and target area overlaps ratio, if >=70%, by the binary system
Label is set as positive number；If≤30%, the binary label is set as negative；Remaining is given up.
Further, the step S40 is specifically included:
Step S41, an image impairment function, a Classification Loss function and a recurrence are defined based on the binary label
Loss function；
Step S42, using the characteristic layer reassigned as the input of normalization exponential function, based on recurrence loss function pair
Candidate region carries out recurrence calculating, based on Classification Loss function according to the classification foundation to return calculate after candidate region into
Row classification, demarcates cervical lesions region based on image impairment function.
Further, in the step S41, described image loss function are as follows:
Wherein i indicates the candidate regions chosen
The index in domain；PiIndicate that candidate region i is the probability of target area；The value of binary label is indicated, if candidate region is mesh
Region is marked, thenOtherwisetiIndicate the coordinate vector of 4 endpoints in candidate region；Indicate the end of real estate 4
The coordinate vector of point；NclsIndicate change parameter when classifying to candidate region；NregCandidate region is normalized in expression
When parameter；λ indicates constant balance factor；
The Classification Loss function are as follows:
The recurrence loss function are as follows:
tx=(x-xa)/wa, ty=(y-ya)/ha, tw=log (w/wa), th=log (h/ha),
Wherein x, y indicate that the centre coordinate of sliding window, w indicate that the width of sliding window, h indicate the height of sliding window；xa,
yaIndicate the centre coordinate of candidate region, waIndicate the width of candidate region, haIndicate the height of candidate region；x*,y*Indicate target area
The centre coordinate in domain, w*Indicate the width of target area, h*Indicate the height of target area.
The present invention has the advantages that
Lesion region is learnt by convolutional neural networks, and each depth is preferably indicated by Squeeze operation
The significance level for spending feature operates enhancing useful feature by Excitation and inhibits unnecessary feature, greatly improves
The screening efficiency of lesion region greatly improves the accuracy in detection of cervical carcinoma lesion region, is conducive to doctor and further examines
Disconnected cervical lesions grade is of great significance for improving uterine neck screening accuracy.
Detailed description of the invention
The present invention is further illustrated in conjunction with the embodiments with reference to the accompanying drawings.
Fig. 1 is a kind of flow chart of cervical lesions method for detecting area of the present invention.
Specific embodiment
It please refers to shown in Fig. 1, a kind of preferred embodiment of cervical lesions method for detecting area of the present invention, including walks as follows
It is rapid:
Step S10, uterine neck image is shot by gynecatoptron and is sent to computer；
Step S20, received image is pre-processed；
Step S30, using region candidate network (RNP network) by choosing candidate region on pretreated image；
Step S40, recurrence calculating is carried out to candidate region using normalization exponential function (Softmax function), and then right
Cervical lesions region is demarcated.
The step S20 is specifically included:
Step S21, received image is normalized；Normalization is a kind of mode of simplified calculating, i.e., will have
The expression formula of dimension turns to nondimensional expression formula, becomes scalar by transformation.
Step S22, depth characteristic is extracted from the image after normalized.The depth characteristic is convolutional neural networks
The middle feature after multiple convolution and pondization operation.
The step S21 is specifically included:
Step S211, the RGB image for being N*N at size by received image scaling, wherein N > 0；
Step S212, by standard data set Software for producing (labelImg), cervix disease is carried out to the RGB image after scaling
Become area marking, and output format is the labeled data of VOC data set (target detection common data sets)；
Step S213, the image input convolutional neural networks of mark are created into training pattern.
The step S22 is specifically included:
Step S221, classification foundation of the shallow-layer feature of the training pattern as lesion region is chosen；The shallow-layer is special
Sign is the feature of convolutional neural networks three first layers convolution Chi Huahou；
Step S222, depth characteristic is extracted to the training pattern:
WhereinIndicate will own in convolutional neural networks
The Feature Mapping that residual error module generates is as input, WLAnd WL-1Successively indicate that two 3 × 3 convolution weight matrixs, BN () indicate
The output data of each hidden layer of convolutional neural networks is normalized, f () indicates ReLU activation primitive,It indicates
The convolution operation of convolutional neural networks；
Step S223, the global draw pond operated by Squeeze, presses the depth characteristic after convolution operation
Contracting, so that the real number ordered series of numbers of C characteristic layer boil down to 1*1*C, in order to preferably indicate the significance level of each feature, a reality
The corresponding significance level for indicating one layer of feature of number:
Wherein C indicates that the port number of characteristic layer, W indicate the width of characteristic layer,
H indicates the height of characteristic layer, ucIndicate characteristic layer channel, i, j are positive integer, Fsq() indicates Squeeze operation；The characteristic layer
Indicate the weight matrix that convolution algorithm generates；
Step S224, it is operated by Excitation by ZcWeight as characteristic layer carries out weight to every layer of depth characteristic
Distribution:
Fex(Zc, W) and=σ (W2δ(W1Zc)), wherein Fex() indicates Excitation operation, and δ indicates ReLU activation primitive, W1
Indicate the parameter that full connection generates for the first time, W2Indicate the parameter that second of full connection generates.Full connection is indicated every characteristic layer
Each node be connected with all nodes of a upper characteristic layer, for all characteristic synthetics to be got up.
Squeeze operation and Excitation operation belong to SE module (Squeeze-and-Excitation
Networks function), SE module are adaptively recalibrated logical by explicitly modeling the relation of interdependence between channel
The characteristic response of road formula.
The step S30 is specifically included:
Step S31, by any scale image input area candidate network；
Step S32, described any scale image generates characteristic pattern using the convolution inclusion layer of convolutional neural networks；The volume
Product inclusion layer is the characteristic layer that convolutional neural networks generate, classification and recurrence while using this characteristic layer, so crying altogether
Enjoy layer；
Step S33, multiple dimensioned convolution operation is carried out on the characteristic pattern and chooses candidate region, and gives each candidate region
Distribution one for mark whether be lesion region binary label；
Step S34, the set of region candidate network output candidate region.
The step S33 is specifically included:
Step S331, sliding carries out the sliding window of selection feature at random on characteristic pattern for creation one；
Step S332, centered on the center of the sliding window, using 3 kinds of scales and 3 kinds of length-width ratios on characteristic pattern
Map the candidate region of 9 kinds of different scales；
Step S333, to each candidate region distribution one for mark whether be lesion region binary label；
Step S334, judge that the I OU of candidate region and target area overlaps ratio, if >=70%, by the binary system
Label is set as positive number；If≤30%, the binary label is set as negative；Remaining is given up, because its to whether diseased region
The classification in domain does not help.
The step S40 is specifically included:
Step S41, an image impairment function, a Classification Loss function and a recurrence are defined based on the binary label
Loss function；
Step S42, it using the characteristic layer reassigned as the input of normalization exponential function (classifier), is damaged based on returning
It loses function and recurrence calculating is carried out to candidate region, based on Classification Loss function according to the classification foundation to the time returned after calculating
Favored area is classified, and is demarcated based on image impairment function to cervical lesions region.
In the step S41, described image loss function are as follows:
Wherein i indicates the candidate regions chosen
The index in domain；PiIndicate that candidate region i is the probability of target area；The value of binary label is indicated, if candidate region is mesh
Region is marked, thenOtherwisetiIndicate the coordinate vector of 4 endpoints in candidate region；Indicate the end of real estate 4
The coordinate vector of point；NclsIndicate change parameter when classifying to candidate region；NregCandidate region is normalized in expression
When parameter；λ indicates constant balance factor；
The Classification Loss function are as follows:
The recurrence loss function are as follows:
tx=(x-xa)/wa, ty=(y-ya)/ha, tw=log (w/wa), th=log (h/ha),
Wherein x, y indicate that the centre coordinate of sliding window, w indicate that the width of sliding window, h indicate the height of sliding window；xa,
yaIndicate the centre coordinate of candidate region, waIndicate the width of candidate region, haIndicate the height of candidate region；x*,y*Indicate target area
The centre coordinate in domain, w*Indicate the width of target area, h*Indicate the height of target area.
In conclusion the present invention has the advantages that
Lesion region is learnt by convolutional neural networks, and each depth is preferably indicated by Squeeze operation
The significance level for spending feature operates enhancing useful feature by Excitation and inhibits unnecessary feature, greatly improves
The screening efficiency of lesion region greatly improves the accuracy in detection of cervical carcinoma lesion region, is conducive to doctor and further examines
Disconnected cervical lesions grade is of great significance for improving uterine neck screening accuracy.
Although specific embodiments of the present invention have been described above, those familiar with the art should be managed
Solution, we are merely exemplary described specific embodiment, rather than for the restriction to the scope of the present invention, it is familiar with this
The technical staff in field should be covered of the invention according to modification and variation equivalent made by spirit of the invention
In scope of the claimed protection.
Claims (8)
1. a kind of cervical lesions method for detecting area, it is characterised in that: described method includes following steps:
Step S10, uterine neck image is shot by gynecatoptron and is sent to computer；
Step S20, received image is pre-processed；
Step S30, using region candidate network by choosing candidate region on pretreated image；
Step S40, recurrence calculating is carried out to candidate region using normalization exponential function, and then cervical lesions region is marked
It is fixed.
2. a kind of cervical lesions method for detecting area as described in claim 1, it is characterised in that: the step S20 is specifically wrapped
It includes:
Step S21, received image is normalized；
Step S22, depth characteristic is extracted from the image after normalized.
3. a kind of cervical lesions method for detecting area as claimed in claim 2, it is characterised in that: the step S21 is specifically wrapped
It includes:
Step S211, the RGB image for being N*N at size by received image scaling, wherein N > 0；
Step S212, by standard data set Software for producing, cervical lesions area marking is carried out to the RGB image after scaling；
Step S213, the image input convolutional neural networks of mark are created into training pattern.
4. a kind of cervical lesions method for detecting area as claimed in claim 3, it is characterised in that: the step S22 is specifically wrapped
It includes:
Step S221, classification foundation of the shallow-layer feature of the training pattern as lesion region is chosen；
Step S222, depth characteristic is extracted to the training pattern:
WhereinIt indicates residual errors all in convolutional neural networks
The Feature Mapping that module generates is as input, WLAnd WL-1Successively indicate that two 3 × 3 convolution weight matrixs, BN () are indicated to volume
The output data of the product each hidden layer of neural network is normalized, and f () indicates ReLU activation primitive,Indicate convolution
The convolution operation of neural network；
Step S223, the global draw pond operated by Squeeze, compresses the depth characteristic after convolution operation, makes
Obtain the real number ordered series of numbers of C characteristic layer boil down to 1*1*C:
Wherein C indicates that the port number of characteristic layer, W indicate the width of characteristic layer, H table
Show the height of characteristic layer, ucIndicate characteristic layer channel, i, j are positive integer, Fsq() indicates Squeeze operation；
Step S224, it is operated by Excitation by ZcWeight as characteristic layer reassigns every layer of depth characteristic:
Fex(Zc, W) and=σ (W2δ(W1Zc)), wherein Fex() indicates Excitation operation, and δ indicates ReLU activation primitive, W1It indicates
The parameter that full connection generates for the first time, W2Indicate the parameter that second of full connection generates.
5. a kind of cervical lesions method for detecting area as described in claim 1, it is characterised in that: the step S30 is specifically wrapped
It includes:
Step S31, by any scale image input area candidate network；
Step S32, described any scale image generates characteristic pattern using the convolution inclusion layer of convolutional neural networks；
Step S33, multiple dimensioned convolution operation is carried out on the characteristic pattern and chooses candidate region, and is distributed to each candidate region
One for mark whether be lesion region binary label；
Step S34, the set of region candidate network output candidate region.
6. a kind of cervical lesions method for detecting area as claimed in claim 5, it is characterised in that: the step S33 is specifically wrapped
It includes:
Step S331, sliding carries out the sliding window of selection feature at random on characteristic pattern for creation one；
Step S332, centered on the center of the sliding window, 9 are mapped on characteristic pattern using 3 kinds of scales and 3 kinds of length-width ratios
The candidate region of kind different scale；
Step S333, to each candidate region distribution one for mark whether be lesion region binary label；
Step S334, judge that the IOU of candidate region and target area overlaps ratio, if >=70%, by the binary label
It is set as positive number；If≤30%, the binary label is set as negative；Remaining is given up.
7. a kind of cervical lesions method for detecting area as claimed in claim 6, it is characterised in that: the step S40 is specifically wrapped
It includes:
Step S41, an image impairment function, a Classification Loss function and a recurrence loss are defined based on the binary label
Function；
Step S42, using the characteristic layer reassigned as the input of normalization exponential function, based on recurrence loss function to candidate
Region carries out recurrence calculating, is divided according to the classification foundation the candidate region returned after calculating based on Classification Loss function
Class demarcates cervical lesions region based on image impairment function.
8. a kind of cervical lesions method for detecting area as claimed in claim 7, it is characterised in that: described in the step S41
Image impairment function are as follows:
Wherein i indicates the rope for the candidate region chosen
Draw；PiIndicate that candidate region i is the probability of target area；Indicate the value of binary label, if candidate region is target area,
ThenOtherwisetiIndicate the coordinate vector of 4 endpoints in candidate region；Indicate the seat of 4 endpoints in real estate
Mark vector；NclsIndicate change parameter when classifying to candidate region；NregIndicate ginseng when candidate region is normalized
Number；λ indicates constant balance factor；
The Classification Loss function are as follows:
The recurrence loss function are as follows:
tx=(x-xa)/wa, ty=(y-ya)/ha, tw=log (w/wa), th=log (h/ha),
Wherein x, y indicate that the centre coordinate of sliding window, w indicate that the width of sliding window, h indicate the height of sliding window；xa,yaTable
Show the centre coordinate of candidate region, waIndicate the width of candidate region, haIndicate the height of candidate region；x*,y*Indicate target area
Centre coordinate, w*Indicate the width of target area, h*Indicate the height of target area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910602980.5A CN110335267A (en) | 2019-07-05 | 2019-07-05 | A kind of cervical lesions method for detecting area |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910602980.5A CN110335267A (en) | 2019-07-05 | 2019-07-05 | A kind of cervical lesions method for detecting area |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110335267A true CN110335267A (en) | 2019-10-15 |
Family
ID=68144293
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910602980.5A Pending CN110335267A (en) | 2019-07-05 | 2019-07-05 | A kind of cervical lesions method for detecting area |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110335267A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111415331A (en) * | 2020-03-03 | 2020-07-14 | 北京深睿博联科技有限责任公司 | Abnormality detection method and system based on category relation in positive chest radiograph |
| CN111415350A (en) * | 2020-03-27 | 2020-07-14 | 福建省妇幼保健院 | Colposcope image identification method for detecting cervical lesions |
| CN112200253A (en) * | 2020-10-16 | 2021-01-08 | 武汉呵尔医疗科技发展有限公司 | Cervical cell image classification method based on senet |
| CN112614099A (en) * | 2020-12-17 | 2021-04-06 | 杭州电子科技大学 | Cervical cancer lesion region detection method based on fast-RCNN model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106340016A (en) * | 2016-08-31 | 2017-01-18 | 湖南品信生物工程有限公司 | DNA quantitative analysis method based on cell microscope image |
| US20170206431A1 (en) * | 2016-01-20 | 2017-07-20 | Microsoft Technology Licensing, Llc | Object detection and classification in images |
| CN109190540A (en) * | 2018-06-06 | 2019-01-11 | 腾讯科技（深圳）有限公司 | Biopsy regions prediction technique, image-recognizing method, device and storage medium |
| CN109636805A (en) * | 2018-11-19 | 2019-04-16 | 浙江大学山东工业技术研究院 | A kind of uterine neck image lesion region segmenting device and method based on classification priori |
-
2019
- 2019-07-05 CN CN201910602980.5A patent/CN110335267A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170206431A1 (en) * | 2016-01-20 | 2017-07-20 | Microsoft Technology Licensing, Llc | Object detection and classification in images |
| CN106340016A (en) * | 2016-08-31 | 2017-01-18 | 湖南品信生物工程有限公司 | DNA quantitative analysis method based on cell microscope image |
| CN109190540A (en) * | 2018-06-06 | 2019-01-11 | 腾讯科技（深圳）有限公司 | Biopsy regions prediction technique, image-recognizing method, device and storage medium |
| CN109636805A (en) * | 2018-11-19 | 2019-04-16 | 浙江大学山东工业技术研究院 | A kind of uterine neck image lesion region segmenting device and method based on classification priori |
Non-Patent Citations (2)
| Title |
|---|
| JIE HU 等: "Squeeze-and-Excitation Networks", 《2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| 范伟康: "基于改进 Faster R-CNN 的肺结节检测", 《中国优秀硕士学位论文全文数据库 卫生医药科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111415331A (en) * | 2020-03-03 | 2020-07-14 | 北京深睿博联科技有限责任公司 | Abnormality detection method and system based on category relation in positive chest radiograph |
| CN111415331B (en) * | 2020-03-03 | 2023-05-23 | 北京深睿博联科技有限责任公司 | Abnormal detection method and system based on category relation in positive chest radiography |
| CN111415350A (en) * | 2020-03-27 | 2020-07-14 | 福建省妇幼保健院 | Colposcope image identification method for detecting cervical lesions |
| CN111415350B (en) * | 2020-03-27 | 2023-04-07 | 福建省妇幼保健院 | Colposcope image identification method for detecting cervical lesions |
| CN112200253A (en) * | 2020-10-16 | 2021-01-08 | 武汉呵尔医疗科技发展有限公司 | Cervical cell image classification method based on senet |
| CN112614099A (en) * | 2020-12-17 | 2021-04-06 | 杭州电子科技大学 | Cervical cancer lesion region detection method based on fast-RCNN model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110335267A (en) | A kind of cervical lesions method for detecting area | |
| Li et al. | Joint multiple fully connected convolutional neural network with extreme learning machine for hepatocellular carcinoma nuclei grading | |
| CN108305249A (en) | The quick diagnosis and methods of marking of full size pathological section based on deep learning | |
| Chan et al. | Texture-map-based branch-collaborative network for oral cancer detection | |
| CN111429407B (en) | Chest X-ray disease detection device and method based on double-channel separation network | |
| CN103096786A (en) | Image analysis for cervical neoplasia detection and diagnosis | |
| CN112184659A (en) | Lung image processing method, device and equipment | |
| JP7312510B1 (en) | Whole-slide pathological image classification system and construction method considering tumor microenvironment | |
| CN111369501B (en) | Deep learning method for identifying oral squamous cell carcinoma based on visual features | |
| Li et al. | Mass detection in mammograms by bilateral analysis using convolution neural network | |
| CN114677378B (en) | Computer-aided diagnosis and treatment system based on ovarian tumor benign and malignant prediction model | |
| Xu et al. | Mammographic mass segmentation using multichannel and multiscale fully convolutional networks | |
| Yuan et al. | Pulmonary nodule detection using 3-d residual u-net oriented context-guided attention and multi-branch classification network | |
| CN109871869A (en) | A kind of Lung neoplasm classification method and its device | |
| Yonekura et al. | Glioblastoma multiforme tissue histopathology images based disease stage classification with deep CNN | |
| CN101968851B (en) | Medical image processing method based on dictionary studying upsampling | |
| Azour et al. | An efficient transfer and ensemble learning based computer aided breast abnormality diagnosis system | |
| Fan et al. | Colposcopic multimodal fusion for the classification of cervical lesions | |
| CN110399899A (en) | Uterine neck OCT image classification method based on capsule network | |
| Abd-Alhalem et al. | Cervical cancer classification based on a bilinear convolutional neural network approach and random projection | |
| CN112017772B (en) | Method and system for constructing disease cognitive model based on female leucorrhea | |
| Xu et al. | A High‐Precision Classification Method of Mammary Cancer Based on Improved DenseNet Driven by an Attention Mechanism | |
| CN116580017A (en) | Improved Mask-R-CNN lung nodule auxiliary detection method integrating double-path channel attention and cavity space attention | |
| Wang et al. | Controlling False-Positives in Automatic Lung Nodule Detection by Adding 3D Cuboid Attention to a Convolutional Neural Network | |
| Li et al. | MC‐UNet: Multimodule Concatenation Based on U‐Shape Network for Retinal Blood Vessels Segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191015 |
|
| RJ01 | Rejection of invention patent application after publication |