CN110473243A - Tooth dividing method, device and computer equipment based on depth profile perception - Google Patents
Tooth dividing method, device and computer equipment based on depth profile perception Download PDFInfo
- Publication number
- CN110473243A CN110473243A CN201910733040.XA CN201910733040A CN110473243A CN 110473243 A CN110473243 A CN 110473243A CN 201910733040 A CN201910733040 A CN 201910733040A CN 110473243 A CN110473243 A CN 110473243A
- Authority
- CN
- China
- Prior art keywords
- profile
- dental imaging
- depth profile
- mask
- tooth
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 230000008447 perception Effects 0.000 title claims abstract description 18
- 230000011218 segmentation Effects 0.000 claims abstract description 45
- 238000012545 processing Methods 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000003384 imaging method Methods 0.000 claims description 71
- 230000006870 function Effects 0.000 claims description 26
- 238000012549 training Methods 0.000 claims description 26
- 230000000877 morphologic effect Effects 0.000 claims description 13
- 230000017105 transposition Effects 0.000 claims description 12
- 238000003708 edge detection Methods 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 5
- 238000004891 communication Methods 0.000 claims description 3
- 235000009508 confectionery Nutrition 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000004804 winding Methods 0.000 claims description 2
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 238000012360 testing method Methods 0.000 abstract description 2
- 230000000007 visual effect Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 238000003709 image segmentation Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000000214 mouth Anatomy 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- 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
- G06T2207/30036—Dental; Teeth
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (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 invention belongs to Medical Image Processings and technical field of computer vision, are related to a kind of tooth dividing method, device and computer equipment based on depth profile perception.The dividing method includes carrying out pretreatment operation to the picture that data are concentrated, and original mask therein is extracted profile mask, and carries out data expansion process to overstriking profile；Using profile mask as supervision message, makes original image by full convolutional network, obtain contour prediction probability graph；Construct U-shaped depth profile sensing network；Using original mask as supervision message, through U-shaped depth profile sensing network after merging original image and contour prediction probability graph, segmentation network is obtained；The original image of test set is sent into trained U-shape network, obtains tooth segmentation result, and be more clear segmentation result with disk filter.Invention increases profile informations, so that improving the accuracy rate of segmentation and the visual effect of segmentation under tooth and the unsharp situation of tooth surrounding tissue boundary.
Description
Technical field
The invention belongs to Medical Image Processing and technical field of computer vision, it is related to a kind of based on depth profile perception
Tooth dividing method, device and computer equipment.
Background technique
In dental field, iconography image is the basic data source of auxiliary diagnosis, x-ray image in dental medicine by with
To check the situation of tooth, gum, the jawbone and skeletal structure in oral cavity etc..Tooth segmentation result may be widely used for tooth just
The scenes such as abnormal, tooth body plantation and forensic identification.
In some cases since x-ray imaging effect limits, such as tooth and small (the especially tooth of surrounding tissue contrast
Root zone domain) cause obscure boundary clear, space problem present in agomphosis, imaging film noise is more, and the filling of tooth body causes metal
Artifact etc. divides tooth for the later period automatically and brings great challenge.
The conventional method of medical image segmentation, can be divided into five major class, be the dividing method based on threshold value respectively, be based on side
The dividing method of edge, the dividing method based on region, the dividing method based on cluster and the dividing method based on watershed.It is based on
The dividing method of threshold value generally only considered the gray value of pixel itself, not account for space characteristics, thus very sensitive to noise；
The contradiction between noise immunity and detection accuracy when dividing method based on edge is difficult to solve edge detection, will lead to and not conform to
The profile of reason and the deviation of position；Dividing method based on region be easy to cause the over-segmentation of image；Segmentation based on cluster
Method depends on the selection of cluster centre, and result may be made to deviate global optimum；Based on the dividing method in watershed to faint
The phenomenon that there is good response at edge, but the noise in image can make watershed algorithm generate over-segmentation.
In recent years, to look for that achieve in medical image segmentation task based on the image analysis method of deep learning good
Progress, has obtained the extensive concern of medical domain.
Currently, the tooth segmentation task based on deep learning is primarily present two big challenges:
(1) tooth to contact with each other is difficult to divide.First is that contacting with each other between tooth, so that the boundary of tooth is difficult to boundary
It is fixed；Second is that tooth and surrounding tissue contrast are small, and noise is more, and in crown portion due to the imaging of tooth pathological image
There may be metal artifacts for position, influence last segmentation effect.
(2) spatial information is lost.The receptive field of shallow-layer network is smaller, due to the limitation of receptive field, can only focus on office
Portion's information is not easy to that global information is combined to obtain the segmentation for meeting tooth space structure.
Summary of the invention
In view of this, the purpose of the present invention is to provide by tooth dividing method that depth profile perceives, device and based on
Calculate machine equipment more particularly to a kind of tooth dividing method based on full convolutional network and U-shaped depth profile sensing network.It is wherein complete
Convolutional network is used for predicted teeth profile diagram, and provides additional edge for U-shaped depth profile sensing network and assist predictive information；
And the depth profile sensing network with U-shaped structure, original graph can be directly upsampled in each unit addition of up-sampling
As the transposition convolutional layer of size, the loss for the boundary characteristic being added can be prevented, due to introducing fine edge auxiliary information,
The image segmentation result of the sharpness of border of available pixel scale.
A kind of tooth dividing method based on depth profile perception of the invention, the described method comprises the following steps:
S1, dental imaging data set is obtained, pretreatment operation is carried out to dental imaging therein, and as training set
Image；
S2, the two-value original mask manually marked from training set image extract profile mask simultaneously by Morphological scale-space
By its overstriking；
S3, using the profile mask after overstriking as the first supervision message, by pretreated original dental imaging by complete
Convolutional network, minimizes first-loss function, and the training full convolutional network obtains contour prediction probability graph；
S4, the U-shaped depth profile sensing network including constricted path and path expander is constructed；
S5, using the original mask as the second supervision message, make pretreated dental imaging and contour prediction probability
Figure is merged, and by U-shaped depth profile sensing network after fusion, obtains tooth segmentation result figure, by minimizing the second damage
Lose function, the training U-shaped depth profile sensing network；
Dental imaging to be split is carried out pretreatment identical with step S1 by S6, the dental imaging to be split for obtaining shooting
Pretreated dental imaging to be predicted is sent into trained U-shaped depth profile sensing network, obtains tooth to be split by operation
The coarse segmentation result of tooth image；
S7, the segmentation result is smoothed, obtains the thin segmentation result of dental imaging.
Optionally, dental imaging data set includes but is not limited to the dental imaging of x-ray shooting, other can divide nature
Image and medical image can also be used as Tooth image data set of the present invention；
Preferably, pretreatment operation can be that original dental imaging is converted to uniform sizes；Its size can for 512 ×
1024,600 × 800 etc..
Further, including the edge of the two-value original mask manually marked out using candy operator extraction, profile is obtained
Exposure mask handles profile mask using data expansion, its profile of overstriking, wherein data expansion processing includes being added with disk filter
Coarse contour.
Wherein, following procedure realization can be used in candy operator:
1) noise is filtered out with smoothed image using Gaussian filter.
2) gradient intensity of each pixel and direction in image are calculated.
3) application non-maximum (Non-Maximum Suppression) inhibits, spuious to eliminate edge detection bring
Response.
4) it detects using dual threshold (Double-Threshold) to determine true and potential edge.
5) by inhibiting isolated weak edge to be finally completed edge detection.
Preferably, the operation of data expansion processing includes: to be added the profile mask extracted using disk filter
Bulk processing.
Further, the contour prediction probability graph in step S3 includes by the dental imaging in training set and through form
The profile mask that method is extracted minimizes the acquisition of first-loss function, the first-loss function by full convolutional network
Including cross entropy loss function, calculation formula are as follows:
Wherein, N indicates pixel number,Indicate the prediction to pixel i, y(i)Indicate that pixel i is corresponding true
Label.
Further, the U-shaped depth profile sensing network includes constricted path and path expander, the constricted path packet
5 duplicate units are included, each unit includes 2 convolutional layers, 1 pond layer；The path expander includes and constricted path feature
Identical 5 units of depth, each unit includes 2 convolutional layers and 1 transposition convolutional layer, and will correspond to the constricted path of depth
Unit and path expander unit are contacted, wherein the series winding includes being overlapped characteristic pattern；By minimizing the second loss letter
The number training U-shaped depth profile sensing network.
Optionally, second loss function includes
Wherein, a indicates the segmentation dental imaging predicted；B indicates original mask；Introducing Dice loss function is to mention
Duplication between the segmentation result and true original mask of height prediction.
Further, the smoothing processing namely corrosion process specifically include and handle consistent circle using with data expansion
Disk filter reduces the boundary of the segmentation result of prediction, so that partitioning boundary is more clear.
Optionally, the radius of disk filter is 2 or 2.5 or 3.
The invention also provides a kind of tooth segmenting device based on depth profile perception, described device includes:
Image collection module, for obtaining dental imaging data set and dental imaging to be split；
Morphological scale-space module, for the original mask in training set image to be extracted its profile by Morphological scale-space
Exposure mask；
Contour prediction probabilistic module, for obtaining contour prediction probability graph by profile mask；
Net structure module, for constructing the U-shaped depth profile sensing network including constricted path and path expander；
Image co-registration module, for merging dental imaging and contour prediction probability graph；
Dental imaging to be split is passed through U-shaped depth profile sensing network, obtains dental imaging by image coarse segmentation module
Coarse segmentation result；
Module is cut in image subdivision, and the coarse segmentation result of dental imaging is smoothed, the subdivision of dental imaging is obtained
Cut result.
Further, the Morphological scale-space module includes:
Gray scale processing unit, for colored dental imaging to be carried out gray processing；
Gaussian filter, for the dental imaging after gray processing to be smoothed and denoise；
Edge detection unit, for extracting the profile mask of dental imaging；
Expansion process unit is used for overstriking profile mask.
Further, the contour prediction probabilistic module includes:
Full convolutional network unit, the network unit including full convolutional coding structure, for predicting contour prediction probability graph；
First supervision unit, for using the profile mask after overstriking as the first supervision message；
First-loss function unit, for according to pixel point prediction and the corresponding true tag of pixel, optimization instruction
Practice full convolutional network unit.
Further, the net structure module includes:
Second loss function unit, for according to segmentation dental imaging and original mask the training U-shaped depth wheel predicted
Wide sensing network；
Second supervision unit, for using original mask as the second supervision message；
U-shaped depth profile sensing network, including constricted path layer and path expander layer；
Constricted path layer, including 5 duplicate units, each unit include 2 convolutional layers and 1 pond layer；
Path expander layer, including with consistent 5 units of constricted path depths of features, each unit include two convolutional layers
With a transposition convolutional layer；
The constricted path unit and path expander unit of corresponding depth are connected.
The invention also provides a kind of computer equipments, including at least one processor；And with the processor communication
At least one processor of connection, in which: the memory is stored with the program instruction that can be executed by the processor, the place
Reason device calls described program instruction to be able to carry out method proposed by the present invention.
Beneficial effects of the present invention:
1) the invention proposes a kind of tooth dividing method based on depth profile perception, this method can be with tooth boundary not
In the case of clearly, the accuracy rate of segmentation and the visual effect of segmentation are improved.
2) present invention merges local message and global information, can be improved the biggish cutting object of area coverage
Segmentation precision can be widely applied to the segmentation of biggish tissue or cell.
Detailed description of the invention
In order to keep the purpose of the present invention, technical scheme and beneficial effects clearer, the present invention provides following attached drawing and carries out
Illustrate:
Fig. 1 is overall flow schematic diagram of the present invention；
Fig. 2 is the full convolutional neural networks structural schematic diagram that the present invention uses；
Fig. 3 is the U-shaped depth profile sensing network structural schematic diagram that the present invention uses；
Fig. 4 is data flow schematic diagram of the invention；
Fig. 5 is the tooth prediction result figure that the present invention uses U-shaped depth profile sensing network.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to of the invention real
The technical solution applied in example is clearly and completely described, it is clear that described embodiment is only that present invention a part is implemented
Example, instead of all the embodiments.
As shown in Figure 1, a kind of tooth dividing method based on depth profile perception of this method, may particularly include following step
It is rapid:
S1, dental imaging data set is obtained, pretreatment operation is carried out to dental imaging therein, and as training set
Image；
The two-value original mask that S2 is manually marked from training set image extracts profile mask simultaneously by Morphological scale-space
By its overstriking；
S3, using the profile mask after overstriking as the first supervision message, by pretreated original dental imaging by complete
Convolutional network, minimizes first-loss function, and the training full convolutional network obtains contour prediction probability graph；
In one embodiment, full convolutional network can refer to as shown in Fig. 2, present embodiments providing a kind of with full convolution
The segmentation network of structure, the network include 5 duplicate units, and each unit includes 2 convolutional layers, 1 pond LeakyReLU
Layer, the effect of this pond layer are that some neurons are suppressed in order to prevent, also added BN between convolutional layer and pond layer and return
One change layer, for prevent gradient explode or gradient disappear the problem of.What is be connected directly after 5 duplicate units is 2 convolution
Layer, 2 convolutional layers are used to extract the quantity that more abstract feature does not increase characteristic pattern.Then, by the last one convolutional layer
Output up-sample and 2 times and be added with the output of the 4th pond layer, the result that will add up is added with the output of the 3rd pond layer
It is upsampled to the size of original image again.Wherein, all up-sampling operations are completed with transposition convolutional layer.Obtained network is logical
It crosses and minimizes the available fine segmentation result of first-loss function.
Wherein, the definition of first-loss function is as shown in formula (1):
Wherein, i indicates that some pixel (x, y), N indicate pixel number,Indicate the prediction to pixel i, y(i)
Indicate the corresponding true tag of pixel i.
S4, the U-shaped depth profile sensing network including constricted path and path expander is constructed；
In one embodiment, U-shaped depth profile sensing network can refer to as shown in figure 3, a U-shaped pair is presented in the network
Claim structure, network is made of constricted path and path expander, and constricted path is differentiated for extracting feature, path expander for restoring
Rate.Constricted path is made of 5 duplicate units, and each unit includes 2 convolutional layers, 1 pond LeakyReLU layer, expands road
Diameter is corresponding with unit identical with constricted path depths of features, and each unit includes 2 convolutional layers and 1 transposition convolutional layer, net
Network finally uses softmax activation primitive as pixel class arbiter.In addition, constricted path and path expander character pair is deep
The unit spliced of degree, concatenation are to be overlapped characteristic pattern, so as to combine the characteristic information and location information of image.
Addition can directly be upsampled to the transposition convolutional layer of original image size in each unit of constricted path, and
1st convolutional layer of the last one unit of transposition convolutional layer and path expander is spliced；
In one embodiment, as shown in figure 3, firstly, each unit in constricted path adds transposition convolutional layer, purpose
It is the contour prediction probability graph that additionally adds in order to prevent since its relatively fine feature extracts in down-sampling the process of feature
In fade away.Secondly, by 4 by the output of transposition convolutional layer and the 1st convolutional layer of the last one unit of path expander
Splicing, concatenation is equally to be overlapped characteristic pattern.Finally, by minimizing the second loss function, prediction result and true is improved
Real result degree of overlapping.
Wherein, the definition of the second loss function is as shown in formula (2):
S5, using the original mask as the second supervision message, make pretreated dental imaging and contour prediction probability
Figure is merged, and by U-shaped depth profile sensing network after fusion, obtains tooth segmentation result figure, by minimizing the second damage
Lose function, the training U-shaped depth profile sensing network；
Mixing operation is exactly to carry out characteristic pattern splicing, and probability graph Fusion Features are realized dimensionality reduction by 1 × 1 convolution kernel, drop
The classification of each pixel is predicted after dimension with softmax activation primitive.
As shown in figure 4, in one embodiment, the dental imaging training stage of the present embodiment may include by dental imaging number
Profile is extracted by morphologic expansion process according to the training set image of concentration, profile is passed through into the net with full convolutional coding structure
Network exports profile prediction probability figure, original dental imaging is merged with contour prediction probability graph, and be input to U-shaped depth
In profile sensing network, further according to the second loss function training U-shaped depth profile sensing network.
Dental imaging to be split is carried out pretreatment identical with step S1 by S6, the dental imaging to be split for obtaining shooting
Pretreated dental imaging to be predicted is sent into trained U-shaped depth profile sensing network, obtains tooth to be split by operation
The coarse segmentation result of tooth image；
This process belongs to test phase, which need to only input dental imaging to be predicted and perceive net to U-shaped depth profile
In network, without using profile information again.
S7, the segmentation result is smoothed, the thin segmentation result for obtaining dental imaging is as shown in Figure 5；It can be with
Find out, the method for proposition through the invention can effectively be partitioned into dental imaging.
Using the boundary for the disk filter smoothing prediction segmentation result that radius is 2, it is more clear partitioning boundary.
The invention also provides a kind of tooth segmenting device based on depth profile perception, described device includes:
Image collection module, for obtaining dental imaging data set and dental imaging to be split；
Morphological scale-space module, for the original mask in training set image to be extracted its profile by Morphological scale-space
Exposure mask；
Contour prediction probabilistic module, for obtaining contour prediction probability graph by profile mask；
Net structure module, for constructing the U-shaped depth profile sensing network including constricted path and path expander；
Image co-registration module, for merging dental imaging and contour prediction probability graph；
Dental imaging to be split is passed through U-shaped depth profile sensing network, obtains dental imaging by image coarse segmentation module
Coarse segmentation result；
Module is cut in image subdivision, and the coarse segmentation result of dental imaging is smoothed, the subdivision of dental imaging is obtained
Cut result.
Further, the Morphological scale-space module includes:
Gray scale processing unit, for colored dental imaging to be carried out gray processing；
Gaussian filter, for the dental imaging after gray processing to be smoothed and denoise；
Edge detection unit, for extracting the profile mask of dental imaging；
Expansion process unit is used for overstriking profile mask.
Further, the contour prediction probabilistic module includes:
Full convolutional network unit, the network unit including full convolutional coding structure, for predicting contour prediction probability graph；
First supervision unit, for using the profile mask after overstriking as the first supervision message；
First-loss function unit, for according to pixel point prediction and the corresponding true tag of pixel, optimization instruction
Practice full convolutional network unit.
Further, the net structure module includes:
Second loss function unit, for according to segmentation dental imaging and original mask the training U-shaped depth wheel predicted
Wide sensing network；
Second supervision unit, for using original mask as the second supervision message；
U-shaped depth profile sensing network, including constricted path layer and path expander layer；
Constricted path layer, including 5 duplicate units, each unit include 2 convolutional layers and 1 pond layer；
Path expander layer, including with consistent 5 units of constricted path depths of features, each unit include two convolutional layers
With a transposition convolutional layer；
The constricted path unit and path expander unit of corresponding depth are connected.
The invention also provides a kind of computer equipments, including at least one processor；And with the processor communication
At least one processor of connection, in which: the memory is stored with the program instruction that can be executed by the processor, the place
Reason device calls described program instruction to be able to carry out method proposed by the present invention.
It is, of course, understood that the Partial Feature of method, apparatus and computer equipment can mutually draw in the present invention
With the present invention will not enumerate to save space.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include: ROM, RAM, disk or CD etc..
Embodiment provided above has carried out further detailed description, institute to the object, technical solutions and advantages of the present invention
It should be understood that embodiment provided above is only the preferred embodiment of the present invention, be not intended to limit the invention, it is all
Any modification, equivalent substitution, improvement and etc. made for the present invention, should be included in the present invention within the spirit and principles in the present invention
Protection scope within.
Claims (10)
1. a kind of tooth dividing method based on depth profile perception, which is characterized in that the described method comprises the following steps:
S1, dental imaging data set is obtained, pretreatment operation is carried out to dental imaging therein, and as training set figure
Picture；
S2, the two-value original mask manually marked from training set image by Morphological scale-space extract profile mask and by its
Overstriking；
S3, using the profile mask after overstriking as the first supervision message, pretreated original dental imaging is passed through into full convolution
Network, minimizes first-loss function, and the training full convolutional network obtains contour prediction probability graph；
S4, the U-shaped depth profile sensing network including constricted path and path expander is constructed；
S5, using the original mask as the second supervision message, make pretreated dental imaging and contour prediction probability graph into
Row fusion, by U-shaped depth profile sensing network after fusion, obtains tooth segmentation result figure, by minimizing the second loss letter
Number, the training U-shaped depth profile sensing network；
Dental imaging to be split is carried out pretreatment operation identical with step S1 by S6, the dental imaging to be split for obtaining shooting,
Pretreated dental imaging to be predicted is sent into trained U-shaped depth profile sensing network, tooth figure to be split is obtained
The coarse segmentation result of picture；
S7, the segmentation result is smoothed, obtains the thin segmentation result of dental imaging.
2. a kind of tooth dividing method based on depth profile perception according to claim 1, which is characterized in that the step
Rapid S2 includes the edge of the two-value original mask manually marked out using candy operator extraction, obtains profile mask, covers to profile
Film is handled using data expansion, its profile of overstriking, and wherein data expansion processing includes using disk filter overstriking profile.
3. a kind of tooth dividing method based on depth profile perception according to claim 1, which is characterized in that step S3
In the contour prediction probability graph include by training set dental imaging and through morphological method extract profile mask lead to
Full convolutional network is crossed, and minimizes the acquisition of first-loss function, the first-loss function is cross entropy loss function, is calculated public
Formula are as follows:
Wherein, N indicates pixel number,Indicate the prediction to pixel i, y(i)Indicate the corresponding true tag of pixel i.
4. a kind of tooth dividing method based on depth profile perception according to claim 1, which is characterized in that the U
The constricted path and path expander of shape depth profile sensing network, the constricted path include 5 duplicate units, each unit
Including 2 convolutional layers, 1 pond layer；The path expander includes 5 units identical with constricted path depths of features, Mei Gedan
Member includes 2 convolutional layers and 1 transposition convolutional layer, and the constricted path unit of corresponding depth and path expander unit are gone here and there
Even, wherein the series winding includes being overlapped characteristic pattern；By minimizing the second loss function training U-shaped depth profile sense
Hownet network.
5. a kind of tooth dividing method based on depth profile perception according to claim 1 or 4, which is characterized in that institute
Stating the second loss function includes
Wherein, a indicates the segmentation dental imaging predicted；B indicates original mask.
6. a kind of tooth segmenting device based on depth profile perception, which is characterized in that described device includes:
Image collection module, for obtaining dental imaging data set and dental imaging to be split；
Morphological scale-space module is covered for the original mask in training set image to be extracted its profile by Morphological scale-space
Film；
Contour prediction probabilistic module, for obtaining contour prediction probability graph by profile mask；
Net structure module, for constructing the U-shaped depth profile sensing network including constricted path and path expander；
Image co-registration module, for merging dental imaging and contour prediction probability graph；
Dental imaging to be split is passed through U-shaped depth profile sensing network, obtains the thick of dental imaging by image coarse segmentation module
Segmentation result；
Module is cut in image subdivision, the coarse segmentation result of dental imaging is smoothed, knot is cut in the subdivision for obtaining dental imaging
Fruit.
7. a kind of tooth segmenting device based on depth profile perception according to claim 6, which is characterized in that the shape
State processing module includes:
Gray scale processing unit, for colored dental imaging to be carried out gray processing；
Gaussian filter, for the dental imaging after gray processing to be smoothed and denoise；
Edge detection unit, for extracting the profile mask of dental imaging；
Expansion process unit is used for overstriking profile mask.
8. a kind of tooth segmenting device based on depth profile perception according to claim 6, which is characterized in that the wheel
Wide prediction probability module includes:
Full convolutional network unit, the network unit including full convolutional coding structure, for predicting contour prediction probability graph；
First supervision unit, for using the profile mask after overstriking as the first supervision message；
First-loss function unit, for it is complete to optimize training according to pixel point prediction and the corresponding true tag of pixel
Convolutional network unit.
9. a kind of tooth segmenting device based on depth profile perception according to claim 6, which is characterized in that the net
Network constructing module includes:
Second supervision unit, for using original mask as the second supervision message；
Second loss function unit, for according to contours segmentation probabilistic image and original mask the training U-shaped depth wheel predicted
Wide sensing network；
U-shaped depth profile sensing network, including constricted path layer and path expander layer；
Constricted path layer, including 5 duplicate units, each unit include 2 convolutional layers and 1 pond layer；
Path expander layer, including with consistent 5 units of constricted path depths of features, each unit include two convolutional layers and one
A transposition convolutional layer；
The constricted path unit and path expander unit of corresponding depth are connected.
10. a kind of computer equipment, including at least one processor；And at least one connecting with the processor communication is deposited
Reservoir, in which: the memory is stored with the program instruction that can be executed by the processor, and the processor calls described program
Instruction is able to carry out method as claimed in claim 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910733040.XA CN110473243B (en) | 2019-08-09 | 2019-08-09 | Tooth segmentation method and device based on depth contour perception and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910733040.XA CN110473243B (en) | 2019-08-09 | 2019-08-09 | Tooth segmentation method and device based on depth contour perception and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473243A true CN110473243A (en) | 2019-11-19 |
| CN110473243B CN110473243B (en) | 2021-11-30 |
Family
ID=68510557
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910733040.XA Active CN110473243B (en) | 2019-08-09 | 2019-08-09 | Tooth segmentation method and device based on depth contour perception and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473243B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110992374A (en) * | 2019-11-28 | 2020-04-10 | 杭州趣维科技有限公司 | Hair refined segmentation method and system based on deep learning |
| CN111161287A (en) * | 2019-12-05 | 2020-05-15 | 山东科技大学 | Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning |
| CN111259772A (en) * | 2020-01-13 | 2020-06-09 | 广州虎牙科技有限公司 | Image annotation method, device, equipment and medium |
| CN111563887A (en) * | 2020-04-30 | 2020-08-21 | 北京航空航天大学杭州创新研究院 | Intelligent analysis method and device for oral cavity image |
| CN111784754A (en) * | 2020-07-06 | 2020-10-16 | 浙江得图网络有限公司 | Tooth orthodontic method, device, equipment and storage medium based on computer vision |
| CN111968120A (en) * | 2020-07-15 | 2020-11-20 | 电子科技大学 | Tooth CT image segmentation method for 3D multi-feature fusion |
| CN112085028A (en) * | 2020-08-31 | 2020-12-15 | 浙江工业大学 | Tooth panoramic semantic segmentation method based on feature map disturbance and boundary supervision |
| CN113139977A (en) * | 2021-04-23 | 2021-07-20 | 西安交通大学 | Mouth cavity curve image wisdom tooth segmentation method based on YOLO and U-Net |
| CN113436734A (en) * | 2020-03-23 | 2021-09-24 | 北京好啦科技有限公司 | Tooth health assessment method and device based on face structure positioning and storage medium |
| CN113516784A (en) * | 2021-07-27 | 2021-10-19 | 四川九洲电器集团有限责任公司 | Tooth segmentation modeling method and device |
| CN113822904A (en) * | 2021-09-03 | 2021-12-21 | 上海爱乐慕健康科技有限公司 | Image labeling device and method and readable storage medium |
| CN116563218A (en) * | 2023-03-31 | 2023-08-08 | 北京长木谷医疗科技股份有限公司 | Spine image segmentation method and device based on deep learning and electronic equipment |
| CN118365964A (en) * | 2024-06-14 | 2024-07-19 | 中国人民解放军空军军医大学 | Deep learning-based recognition method for tooth position and periapical periodontitis of oral cavity curved surface fracture layer |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130308846A1 (en) * | 2011-07-21 | 2013-11-21 | Carestream Health, Inc. | Method for teeth segmentation and alignment detection in cbct volume |
| US20160292878A1 (en) * | 2015-03-30 | 2016-10-06 | General Electric Company | Methods and systems for automatic segmentation |
| CN108492286A (en) * | 2018-03-13 | 2018-09-04 | 成都大学 | A kind of medical image cutting method based on the U-shaped convolutional neural networks of binary channel |
| CN109785334A (en) * | 2018-12-17 | 2019-05-21 | 深圳先进技术研究院 | Cardiac magnetic resonance images dividing method, device, terminal device and storage medium |
| CN109816661A (en) * | 2019-03-22 | 2019-05-28 | 电子科技大学 | A kind of tooth CT image partition method based on deep learning |
| CN109903396A (en) * | 2019-03-20 | 2019-06-18 | 洛阳中科信息产业研究院(中科院计算技术研究所洛阳分所) | A kind of tooth three-dimensional model automatic division method based on surface parameterization |
-
2019
- 2019-08-09 CN CN201910733040.XA patent/CN110473243B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130308846A1 (en) * | 2011-07-21 | 2013-11-21 | Carestream Health, Inc. | Method for teeth segmentation and alignment detection in cbct volume |
| US20160292878A1 (en) * | 2015-03-30 | 2016-10-06 | General Electric Company | Methods and systems for automatic segmentation |
| CN108492286A (en) * | 2018-03-13 | 2018-09-04 | 成都大学 | A kind of medical image cutting method based on the U-shaped convolutional neural networks of binary channel |
| CN109785334A (en) * | 2018-12-17 | 2019-05-21 | 深圳先进技术研究院 | Cardiac magnetic resonance images dividing method, device, terminal device and storage medium |
| CN109903396A (en) * | 2019-03-20 | 2019-06-18 | 洛阳中科信息产业研究院(中科院计算技术研究所洛阳分所) | A kind of tooth three-dimensional model automatic division method based on surface parameterization |
| CN109816661A (en) * | 2019-03-22 | 2019-05-28 | 电子科技大学 | A kind of tooth CT image partition method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| OLAF RONNEBERGER ET AL；: "《U-Net: Convolutional Networks for Biomedical Image Segmentation 》", 《ARXIV:1505.04597V1》 * |
| 李惠: "《基于分割和轮廓特征的医学牙齿图像处理算法研究》", 《硕士论文库》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110992374B (en) * | 2019-11-28 | 2023-09-05 | 杭州小影创新科技股份有限公司 | Hair refinement segmentation method and system based on deep learning |
| CN110992374A (en) * | 2019-11-28 | 2020-04-10 | 杭州趣维科技有限公司 | Hair refined segmentation method and system based on deep learning |
| CN111161287A (en) * | 2019-12-05 | 2020-05-15 | 山东科技大学 | Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning |
| CN111259772A (en) * | 2020-01-13 | 2020-06-09 | 广州虎牙科技有限公司 | Image annotation method, device, equipment and medium |
| CN113436734A (en) * | 2020-03-23 | 2021-09-24 | 北京好啦科技有限公司 | Tooth health assessment method and device based on face structure positioning and storage medium |
| CN113436734B (en) * | 2020-03-23 | 2024-03-05 | 北京好啦科技有限公司 | Tooth health assessment method, equipment and storage medium based on face structure positioning |
| CN111563887B (en) * | 2020-04-30 | 2022-04-22 | 北京航空航天大学杭州创新研究院 | Intelligent analysis method and device for oral cavity image |
| CN111563887A (en) * | 2020-04-30 | 2020-08-21 | 北京航空航天大学杭州创新研究院 | Intelligent analysis method and device for oral cavity image |
| CN111784754B (en) * | 2020-07-06 | 2024-01-12 | 浙江得图网络有限公司 | Tooth orthodontic method, device, equipment and storage medium based on computer vision |
| CN111784754A (en) * | 2020-07-06 | 2020-10-16 | 浙江得图网络有限公司 | Tooth orthodontic method, device, equipment and storage medium based on computer vision |
| CN111968120A (en) * | 2020-07-15 | 2020-11-20 | 电子科技大学 | Tooth CT image segmentation method for 3D multi-feature fusion |
| CN111968120B (en) * | 2020-07-15 | 2022-03-15 | 电子科技大学 | Tooth CT image segmentation method for 3D multi-feature fusion |
| CN112085028A (en) * | 2020-08-31 | 2020-12-15 | 浙江工业大学 | Tooth panoramic semantic segmentation method based on feature map disturbance and boundary supervision |
| CN112085028B (en) * | 2020-08-31 | 2024-03-12 | 浙江工业大学 | Tooth full-scene semantic segmentation method based on feature map disturbance and boundary supervision |
| CN113139977A (en) * | 2021-04-23 | 2021-07-20 | 西安交通大学 | Mouth cavity curve image wisdom tooth segmentation method based on YOLO and U-Net |
| CN113139977B (en) * | 2021-04-23 | 2022-12-27 | 西安交通大学 | Mouth cavity curve image wisdom tooth segmentation method based on YOLO and U-Net |
| CN113516784A (en) * | 2021-07-27 | 2021-10-19 | 四川九洲电器集团有限责任公司 | Tooth segmentation modeling method and device |
| CN113516784B (en) * | 2021-07-27 | 2023-05-23 | 四川九洲电器集团有限责任公司 | Tooth segmentation modeling method and device |
| CN113822904A (en) * | 2021-09-03 | 2021-12-21 | 上海爱乐慕健康科技有限公司 | Image labeling device and method and readable storage medium |
| CN113822904B (en) * | 2021-09-03 | 2023-08-08 | 上海爱乐慕健康科技有限公司 | Image labeling device, method and readable storage medium |
| CN116563218A (en) * | 2023-03-31 | 2023-08-08 | 北京长木谷医疗科技股份有限公司 | Spine image segmentation method and device based on deep learning and electronic equipment |
| CN118365964A (en) * | 2024-06-14 | 2024-07-19 | 中国人民解放军空军军医大学 | Deep learning-based recognition method for tooth position and periapical periodontitis of oral cavity curved surface fracture layer |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473243B (en) | 2021-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110473243A (en) | Tooth dividing method, device and computer equipment based on depth profile perception | |
| CN113516784B (en) | Tooth segmentation modeling method and device | |
| CN111598862B (en) | Breast molybdenum target image segmentation method, device, terminal and storage medium | |
| CN107644225A (en) | Pulmonary lesionses recognition methods, device and realization device | |
| EP0945830B1 (en) | Image processing method including multidimensional image segmentation stages and medical imaging apparatus using the same | |
| CN106339998A (en) | Multi-focus image fusion method based on contrast pyramid transformation | |
| CN111915530A (en) | End-to-end-based haze concentration self-adaptive neural network image defogging method | |
| CN106296653A (en) | Brain CT image hemorrhagic areas dividing method based on semi-supervised learning and system | |
| EP0635806A1 (en) | Method for processing digital pictures for locally determining centre and width of objects having the form of bands contrasting with background | |
| CN110765833A (en) | Crowd density estimation method based on deep learning | |
| CN109741335A (en) | Blood vessel OCT image medium vessels wall and the dividing method and device of blood flow area | |
| CN109919915A (en) | Retina fundus image abnormal region detection method and device based on deep learning | |
| CN111815766A (en) | Processing method and system for reconstructing blood vessel three-dimensional model based on 2D-DSA image | |
| CN113554665A (en) | Blood vessel segmentation method and device | |
| CN115937158A (en) | Stomach cancer focus region segmentation method based on layered attention mechanism | |
| CN112700460A (en) | Image segmentation method and system | |
| CN112102259A (en) | Image segmentation algorithm based on boundary guide depth learning | |
| Gupta et al. | A novel finetuned YOLOv8 model for real-time underwater trash detection | |
| CN113313728A (en) | Intracranial artery segmentation method and system | |
| CN117152430A (en) | Image segmentation model establishment method and application based on U-shaped depth supervision model | |
| CN116309902A (en) | Vascular image reconstruction method, vascular image reconstruction device, electronic equipment and storage medium | |
| CN116128799A (en) | Tumor incisal edge recognition system based on hyperspectral image | |
| CN112862785B (en) | CTA image data identification method, device and storage medium | |
| Krishna et al. | MLRNet: Skin lesion segmentation using hybrid Gaussian guided filter with CNN | |
| CN115760754A (en) | Multi-modality MRI image auditory nerve sheath tumor segmentation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240626 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Country or region after: China Address before: 400065 Chongwen Road, Nanshan Street, Nanan District, Chongqing Patentee before: CHONGQING University OF POSTS AND TELECOMMUNICATIONS Country or region before: China |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240628 Address after: Room 1209, 12th Floor, Building 3, Yard 2, Dongshi East Fourth Road, Tongzhou District, Beijing, 101100 Patentee after: Butts (Beijing) Technology Co.,Ltd. Country or region after: China Address before: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee before: Dragon totem Technology (Hefei) Co.,Ltd. Country or region before: China |
|
| TR01 | Transfer of patent right |