CN108171694A - Nodule detection methods, system and equipment based on convolutional neural networks - Google Patents
Nodule detection methods, system and equipment based on convolutional neural networks Download PDFInfo
- Publication number
- CN108171694A CN108171694A CN201711464206.XA CN201711464206A CN108171694A CN 108171694 A CN108171694 A CN 108171694A CN 201711464206 A CN201711464206 A CN 201711464206A CN 108171694 A CN108171694 A CN 108171694A
- Authority
- CN
- China
- Prior art keywords
- tubercle
- information
- convolutional neural
- neural networks
- dimensional data
- 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
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
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
- G06T2207/30064—Lung nodule
Abstract
The invention discloses a kind of nodule detection methods based on convolutional neural networks, including：The three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, obtains the classification information and location information of tubercle；Output category information and location information；This method is compared with the tubercle recognizer based on two dimensional image, three-dimensional data in the present invention includes more rich, comprehensive information, recognition accuracy more higher than two-dimentional tubercle can be obtained, and the tubercle recognizer for avoiding two dimensional image scores the drawbacks of different to same tubercle difference section.The invention also discloses a kind of nodule detection system based on convolutional neural networks, equipment and computer readable storage mediums, have above-mentioned advantageous effect.
Description
Technical field
The present invention relates to complementary medicine diagnostic field, more particularly to a kind of nodule detection side based on convolutional neural networks
Method, system, equipment and computer readable storage medium.
Background technology
Thyroid nodule refers to the lump in thyroid gland, can be moved up and down with swallowing act with thyroid gland, is clinical
Common illness can be caused by Different types of etiopathogenises.Clinically there are many thyroid disease, as thyroid gland retrogression, inflammation, itself
Immune and neoformation etc. may appear as tubercle.Thyroid nodule can be with single-shot, can also be multiple, and multiple nodules compare single-shot
The incidence of tubercle is high, but the incidence of Solitary nodules thyroid cancer is higher.
At present, there are mainly two types of forms for the detection of all kinds of tubercles such as thyroid nodule.First, artificial judgment depends on
The clinical experience of doctor, and diagnosing there are larger subjectivity to tubercle, experience is difficult to effectively compared with the doctor of shortcoming
Diagnosis.Second, it is diagnosed using two dimensional image, the tubercle deep learning based on two dimensional slice does not recognize tubercle comprehensively, suddenly
The three-dimensional nature of tubercle has been omited, has been likely to be obtained when inevitably omitting compared with multi information, therefore same tubercle difference section being judged
Different results；Certain global features of i.e. two-dimentional tubercle diagnosis from certain sections assessment tubercle are more difficult and more unilateral,
Be likely to occur fail to pinpoint a disease in diagnosis, mistaken diagnosis.
Therefore, accuracy, the reliability of the nodule detection based on convolutional neural networks how are improved, is people in the art
Member's technical issues that need to address.
Invention content
The object of the present invention is to provide a kind of nodule detection methods based on convolutional neural networks, system, equipment and calculating
Machine readable storage medium storing program for executing carries out three-dimensional data deep learning analysis using convolutional neural networks, and three-dimensional data includes more
Abundant, comprehensive information, therefore improve accuracy, the reliability of the nodule detection based on convolutional neural networks.
In order to solve the above technical problems, the present invention provides a kind of nodule detection methods based on convolutional neural networks, it is described
Method includes：
The three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, obtains the classification letter of the tubercle
Breath and location information；
Export the classification information and the location information.
Optionally, it is described that the three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, it obtains described
The classification information and location information of tubercle, including：
The characteristic information of the network extraction three-dimensional data is extracted by three-dimensional feature；
Intensive cuboid frame sequence is generated according to the corresponding volume data of the three-dimensional data；
Using classifying, subnet handles the intensive cuboid frame sequence and the characteristic information, obtains the knot
The classification information of section；And/or
The intensive cuboid frame sequence and the characteristic information are handled using subnet is returned, obtain the knot
The cuboid encirclement frame of section；And/or
The cuboid encirclement frame is handled using subnet is divided, obtains the mask information of the tubercle.
Optionally, it is described that intensive cuboid frame sequence is generated according to the corresponding volume data of the three-dimensional data, including：
The three-dimensional data is detected using RPN networks, obtains candidate VOI regions；
Intensive cuboid frame sequence is generated according to the volume data of the corresponding three-dimensional data in the candidate VOI regions.
Optionally, it is described the cuboid encirclement frame to be handled using dividing subnet, obtain the mask of the tubercle
Information, including：
Non-maxima suppression calculating is carried out to cuboid encirclement frame, obtains accurate encirclement frame；
The volume data in the accurate encirclement frame is split using subnet is divided, obtains the mask letter of the tubercle
Breath.
Optionally, the acquisition of the three-dimensional data, including：
The three-dimensional data is obtained using three dimensional ultrasound probe.
Optionally, it is described that the three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, it obtains described
Before the classification information and location information of tubercle, further include：
Resampling is carried out to the three-dimensional data of the tubercle of acquisition；
By the three-dimensional data after each resampling divided by the maximum value in the three-dimensional data of acquisition, and 0.5 is subtracted, obtained
To normalization volume data.
Optionally, the output classification information and the location information, including：
Three-dimensional rendering processing is carried out to the classification information, the location information and the pretreated volume data,
Visualization output treated data.
The present invention also provides a kind of nodule detection system based on convolutional neural networks, the system comprises：
Deep learning module for being analyzed using convolutional neural networks the three-dimensional data of the tubercle of acquisition, is obtained
To the classification information and location information of the tubercle；
Output module, for exporting the classification information and the location information.
Optionally, the output module is specially to visualize the mould for exporting the classification information and the location information
Block.
The present invention also provides a kind of nodule detection equipment based on convolutional neural networks, including：
Processor for being analyzed using convolutional neural networks the three-dimensional data of the tubercle of acquisition, is obtained described
The classification information and location information of tubercle；
Output device, for exporting the classification information and the location information.
The present invention also provides a kind of computer readable storage medium, calculating is stored on the computer readable storage medium
Machine program realizes the tubercle inspection based on convolutional neural networks described in any of the above-described when the computer program is executed by processor
The step of survey method.
A kind of nodule detection methods based on convolutional neural networks provided by the present invention, including：Utilize convolutional Neural net
Network analyzes the three-dimensional data of the tubercle of acquisition, obtains the classification information and location information of tubercle；Described point of output
Category information and the location information.
As it can be seen that this method analyzes three-dimensional data using convolutional neural networks, with being based on two dimension in the prior art
The tubercle recognizer of image is compared, and the three-dimensional data in this method includes more rich, comprehensive information, can obtain than two
The higher recognition accuracy of tubercle is tieed up, and the tubercle recognizer for avoiding two dimensional image scores not to same tubercle difference section
The drawbacks of same；I.e. this method improves accuracy, the reliability of nodule detection.The present invention also provides one kind to be based on convolutional Neural net
Nodule detection system, equipment and the computer readable storage medium of network have above-mentioned advantageous effect, and details are not described herein.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention, for those of ordinary skill in the art, without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
The flow chart of the nodule detection methods based on convolutional neural networks that Fig. 1 is provided by the embodiment of the present invention；
The stream of a kind of specifically nodule detection methods based on convolutional neural networks that Fig. 2 is provided by the embodiment of the present invention
Journey schematic diagram；
The structure diagram of the nodule detection system based on convolutional neural networks that Fig. 3 is provided by the embodiment of the present invention；
The structure diagram of the nodule detection equipment based on convolutional neural networks that Fig. 4 is provided by the embodiment of the present invention.
Specific embodiment
The core of the present invention is to provide a kind of nodule detection methods based on convolutional neural networks, system, equipment and calculating
Machine readable storage medium storing program for executing carries out deep learning analysis to three-dimensional data using convolutional neural networks, improves based on convolutional Neural
Accuracy, the reliability of the nodule detection of network.
Purpose, technical scheme and advantage to make the embodiment of the present invention are clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical solution in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
Part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
All other embodiments obtained without making creative work shall fall within the protection scope of the present invention.
The present embodiment in order to solve in the prior art using two dimensional image carry out tubercle (such as thyroid nodule) detection,
Meeting drain message is more, and there may be disagreements for the judgement of its two dimensional image to different section tubercles, it might therefore cause to miss
The phenomenon that examining.The present embodiment carries out deep learning analysis using convolutional neural networks to three-dimensional data.Specifically please refer to Fig.1,
The flow chart of the nodule detection methods based on convolutional neural networks that Fig. 1 is provided by the embodiment of the present invention；This method can wrap
It includes：
S110, the three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, obtains the tubercle
Classification information and location information.
Wherein, convolutional neural networks are a kind of deep learning algorithms with supervised learning mechanism, utilize the convolutional Neural
Network can carry out three-dimensional localization, segmentation, good pernicious assessment of tubercle etc..The training to convolutional neural networks of early period can be straight
Connect the training that convolutional neural networks are carried out using the three-dimensional data of tubercle as training data.Utilize the convolutional neural networks after training
Pretreated volume data is analyzed, obtains the classification information and location information of tubercle.Under normal circumstances for convolution
Neural metwork training needs that it is made to have classification, detection and segmentation ability, needs convolutional neural networks energy in the present embodiment
It is enough to identify that the tubercle is benign protuberance, Malignant Nodules or non-nodules according to the pretreated volume data of input.Tied
The classification information of section.The present embodiment in order to further improve the accuracy of the nodule detection based on convolutional neural networks, more added with
The feature of tubercle is integrally held conducive to doctor, is conducive to good pernicious assessment, the present embodiment to tubercle and also needs to convolutional Neural net
Network can obtain the location information of tubercle.The present embodiment does not limit the type of the location information of tubercle, can be according to user
Actual demand is trained.Such as can include segmentation information (such as mask templates, that is, mask information of voxel classification) or
It is the location information for returning information (such as tubercle cuboid encirclement frame) etc. or tubercle, returns information and divide letter
The arbitrary combination of breath.It, can be with such as when pretreated volume data being detected, classifies and divided by convolutional neural networks
The three-dimensional mask for obtaining cuboid encirclement frame (testing result), good, Malignant Nodules judgement result (classification results) and tubercle is covered
Film (segmentation result).
The present embodiment does not limit the acquisition modes of the three-dimensional data of specific tubercle, and user can be true according to actual demand
Surely the concrete mode of three-dimensional data is acquired.Such as can be by free arm scanning or three dimensional ultrasound probe.Wherein, when using
When free arm scanning obtains three-dimensional data, free arm in scanning process is needed to keep stablizing, and algorithm is needed to carry out image and is matched
It is accurate.When using three dimensional ultrasound probe, three-dimensional data (i.e. B ultrasound volume data) can be directly acquired；Therefore three dimensional ultrasound probe
More stablize compared to free arm scanning, is quick, is accurate.Corresponding subsequent convolutional neural networks can obtain more accurate knot
Fruit.It is preferred, therefore, that three-dimensional data is obtained using three dimensional ultrasound probe in the present embodiment.
Further, the present embodiment, can be to inputting convolutional Neural net in order to improve the accuracy of convolutional neural networks detection
The three-dimensional data of network is pre-processed, so as to improve the reliability of the three-dimensional data of input convolutional neural networks.It is excellent
Choosing, it can also include pre-processing the three-dimensional data of the tubercle of acquisition before step S110, after obtaining pretreatment
Volume data.The present embodiment does not limit the specific mode pre-processed, and user can be selected according to actual conditions.
Wherein, the main purpose of pretreatment is in order to which the three-dimensional data of the tubercle to acquisition pre-processes, so as to protect
The reliability of data analyzed using convolutional neural networks is demonstrate,proved, to improve the accuracy of final detection result.Wherein, it is three-dimensional
Volume data contains all ultrasound datas of solid space where tubercle, and is not only the 2-D data of a certain section, therefore information
It is more abundant comprehensive.Drain message is avoided, can intelligent diagnostics more accurately be carried out to tubercle.I.e. with the knot based on two dimensional image
Section deep learning algorithm (such as convolutional neural networks) is compared, and is carried out network training using two dimensional image, is ignored the three-dimensional of tubercle
Essence, tubercle three-dimensional data include more rich information, can obtain recognition accuracy more higher than two-dimentional tubercle, and avoid
The drawbacks of two-dimentional algorithm is different to the scoring of same tubercle difference section.
The present embodiment does not limit specific pretreatment mode.User can be according to the practical computing capability of hardware and number
The determining corresponding preprocessing process such as the source according to acquisition.Under normal circumstances, pretreatment can include：Resampling, denoising, removal
Corrupt data etc..Therefore, optionally, the three-dimensional data of the tubercle of acquisition is pre-processed, obtains pretreated body
Data can include：
Resampling is carried out to the three-dimensional data of the tubercle of acquisition；
By the three-dimensional data after each resampling divided by the maximum value in the three-dimensional data of acquisition, and 0.5 is subtracted, obtained
To normalization volume data.I.e. pretreated volume data is the normalization volume data of same size.
Specifically, the three-dimensional data obtained to free arm scanning or three dimensional ultrasound probe carries out resampling, sampled
To suitable size.Such as the down-sampled of equal proportion is needed for larger three-dimensional data, and smaller three-dimensional data is then
It carries out equal proportion and rises sampling.The size of resampling can be set or be changed by user.
By the three-dimensional data after each resampling divided by the maximum value in the three-dimensional data of acquisition, and 0.5 is subtracted, obtained
To normalization volume data, be conducive to be promoted the detection of follow-up convolutional neural networks after pretreatment using the normalization volume data obtained
Accuracy.
Complete information in order to obtain, reliably diagnoses tubercle convenient for doctor.Convolution is being used in the present embodiment
When neural network handles pretreated volume data, needing loading network weight file, (wherein, network weight file is
Record the file of convolutional neural networks parameter.Convolutional neural networks need to be loaded into this file, and the present embodiment is not to the network weight
The concrete numerical value of each convolutional neural networks parameter and type are defined in value file, according to the content and reality of subsequent analysis
Border hardware computing capability is selected and is changed).The present embodiment is not defined executive agent, can be processor,
Can be graphics processor (GPU).Preferably, pretreated volume data is analyzed using convolutional neural networks, obtained
The classification information and location information of tubercle can include：
It (can certainly be initial said three-dimensional body number to extract network to extract pretreated volume data by three-dimensional feature
According to) characteristic information；
Wherein, pretreated volume data is input in convolutional neural networks (i.e. CNN networks).In convolutional neural networks
In, high layer information, that is, characteristic information of volume data is extracted first with three-dimensional feature extraction network to obtain the smaller pumping of data volume
As feature (not limiting specific three-dimensional feature extraction network algorithm, such as Resnet3D in the present embodiment).
Intensive cuboid frame sequence is generated according to the corresponding volume data of pretreated volume data；
Wherein, to the corresponding volume data of pretreated volume data, i.e., the length number of pretreated volume data
Different size, different location, a high proportion of intensive cuboid frame sequence for spreading all over volume data of different length and width are generated according to processing is carried out.
Using classifying, subnet handles intensive cuboid frame sequence and characteristic information, obtains the classification letter of tubercle
Breath；And/or
Wherein, classification subnet judge that the volume data part of intensive cuboid frame sequence institute frame choosing is to belong to benign knot
Section, Malignant Nodules or non-nodules.I.e. classification information includes benign protuberance, Malignant Nodules and non-nodules.It further can be with
Provide the probability value of each case.
Intensive cuboid frame sequence and characteristic information are handled using subnet is returned, obtain the cuboid packet of tubercle
Peripheral frame；And/or
Wherein, to different size, different location, a high proportion of intensive cuboid frame sequence for spreading all over volume data of different length and width
Classified, regression training, obtain accurate cuboid encirclement frame.It, can also be in life further for data processing amount is reduced
Data are screened to reduce data volume before into intensive cuboid frame sequence.The present embodiment does not limit specific data sieve
Preferred form of this, such as carry out data screening using RPN networks.It is i.e. preferred, RPN networks can be utilized in the present embodiment to described three
Dimension volume data is detected, and is obtained candidate VOI regions (Volume Of Interest, three dimensional area of interest)；According to described
The volume data of the corresponding three-dimensional data in candidate VOI regions generates intensive cuboid frame sequence.Specifically, to intensive cuboid
Cuboid frame in frame sequence carries out small position movement and the adjustment of length, width and height, and it is more accurate can to obtain target in frame
Position frame.Subnet is returned to fine-tune candidate VOI positions, the starting point that can calculate more accurate candidate VOI is sat
It is marked with and length, width and height, obtains cuboid encirclement frame.Doctor can pay close attention to the region.
The cuboid encirclement frame is handled using subnet is divided, obtains the mask information of the tubercle.
Wherein, for segmentation subnet to divide tubercle, output is to represent the mask templates of voxel classification.That is mask template,
The individual data items block that mask is made of 0,1 or other integers, it is equal in magnitude with pretreated volume data.In this implementation
In example, 0 indicates it belong to background, i.e. non-nodules region, and 1 indicates it belong to benign protuberance region, and 2 indicate it belong to Malignant Nodules
Region.Mask template sizes can be M*M*M in the present embodiment, correspond to each VOI voxel respectively, and the value of voxel is represented for 0
Non-nodules, benign protuberance is represented for 1, and Malignant Nodules are represented for 2.The visualization that tubercle is carried out by the mask information of tubercle can be with
Doctor is facilitated to hold tubercle global feature, promotes diagnosis efficiency and accuracy.
Detailed process can be：
Non-maxima suppression calculating is carried out to the cuboid encirclement frame of tubercle, obtains accurate encirclement frame；So as to effect
Poor recurrence subnet treated cuboid encirclement frame, obtains more accurate cuboid encirclement frame.
The volume data in more accurate cuboid encirclement frame obtained after calculating non-maxima suppression is sent into segmentation
Net is split, and obtains the mask information i.e. Mask masks of the tubercle of target.
S120, output category information and location information.
Specifically, reference can be provided for the doctor of part limited experience by output category information and location information,
Doctor can be allowed to pay close attention to high risk zone, also can prevent from failing to pinpoint a disease in diagnosis as the supplement of diagnosis, doctor is facilitated to hold tubercle whole
Body characteristics promote diagnosis efficiency and accuracy.Unification, the higher tubercle diagnostic result of accuracy can be provided.
The present embodiment does not limit specific output form, such as can be exported using display device, can also
It is word output, naturally it is also possible to be that a variety of output forms combine.Such as display device output position information, and pass through voice and set
It is standby to export the corresponding classification information of the location information.
Based on above-mentioned technical proposal, nodule detection methods based on convolutional neural networks that the embodiment of the present invention carries utilize
Convolutional neural networks analyze three-dimensional data, compared with the tubercle recognizer in the prior art based on two dimensional image,
Three-dimensional data in this method includes more rich, comprehensive information, can obtain recognition accuracy more higher than two-dimentional tubercle,
And the tubercle recognizer for avoiding two dimensional image scores the drawbacks of different to same tubercle difference section；I.e. this method improves base
Accuracy, reliability in the nodule detection of convolutional neural networks.
Based on above-described embodiment, output category information and location information can include：
Visualize output category information and location information.
Specifically, since the nodule detection result based on convolutional neural networks contains location information (i.e. segmentation information),
Therefore if only word output can reduce the efficiency that doctor obtains testing result.It can doctor couple by visualizing output
There is more intuitive impression in specific form of tubercle etc., more quickly capture keynote message, be conducive to doctor's entirety
Tubercle feature is held, is conducive to the good pernicious assessment of tubercle.Such as by visualize output can make doctor see nodular morphology,
Whether the finishing of tubercle edge, the information such as tubercle calcification pattern, and these information are not easy to obtain very by other way of outputs
To being completely to obtain very much.
Further for the display effect for the data for improving visualization output, three-dimensional visualization output can be carried out.It is i.e. excellent
Choosing, visualizing output category information and location information can include：
Three-dimensional rendering processing is carried out to the segmentation information in classification information, location information and pretreated volume data,
Visualization output treated data.
Specifically, after the good information such as classification informations and mask such as pernicious are obtained, three-dimensional wash with watercolours is carried out together with volume data
Dye output.Whether three-dimensional visualization is easier the global feature such as nodular morphology, the finishing of tubercle edge for certain tubercles, tie
Section calcification pattern etc. is understood.The specific execution flowchart process of the present embodiment can be with reference chart 2.It can be with root in the present embodiment
360 ° of displayings are carried out to nodule image according to the demand of doctor, you can be exported under the angle according to the viewing angle that user selects
The tubercle of three-dimensional rendering.
Further, the tubercle recognized is comprehensively observed in order to easily facilitate doctor, the present embodiment can be with
According to instruction input by user, the two dimensional slice image of the tubercle of the selected section of output user.Here it does not limit specific
Extraction two dimensional slice image specific algorithm.Such as can be that interpolation method carries out section extraction.
Based on above-mentioned technical proposal, nodule detection methods based on convolutional neural networks that the embodiment of the present invention carries, to knowing
The tubercle not gone out carries out three-dimensional visualization, and doctor can adjust the knot of viewing angle three-dimensional rendering from any direction manually
Section, and can show the tubercle two dimensional slice image of arbitrary tangent as needed, therefore, doctor can have to tubercle more comprehensively,
Three-dimensional cognition.Further improve accuracy, the reliability of the nodule detection based on convolutional neural networks.
It below can to the nodule detection system provided in an embodiment of the present invention based on convolutional neural networks, equipment and computer
It reads storage medium to be introduced, the nodule detection system described below based on convolutional neural networks, equipment and computer-readable
Storage medium can correspond reference with the above-described nodule detection methods based on convolutional neural networks.
It please refers to Fig.3, the knot of the nodule detection system based on convolutional neural networks that Fig. 3 is provided by the embodiment of the present invention
Structure block diagram；The system can include：
Deep learning module 100, for being analyzed using convolutional neural networks the three-dimensional data of the tubercle of acquisition,
Obtain the classification information and location information of the tubercle；
Output module 200, for exporting the classification information and the location information.
It should be noted that based on above-mentioned any embodiment, the nodule detection system based on convolutional neural networks can be
It is realized based on programmable logic device, programmable logic device includes FPGA, CPLD, microcontroller etc..
Based on above-described embodiment, output module 200 is specially to visualize to export the classification information and position letter
The module of breath.
It please refers to Fig.4, the knot of the nodule detection equipment based on convolutional neural networks that Fig. 4 is provided by the embodiment of the present invention
Structure block diagram；The equipment can include：
Processor 400 for being analyzed using convolutional neural networks the three-dimensional data of the node of acquisition, obtains institute
State the classification information and location information of tubercle；
Output device 500, for exporting the classification information and the location information.
The present embodiment is not defined the form of output device 500.Further, the output device 500 can wrap
Include display device, such as display screen or display etc..
The embodiment of the present invention also provides a kind of computer readable storage medium, and meter is stored on computer readable storage medium
Calculation machine program realizes the tubercle inspection based on convolutional neural networks in above-mentioned any embodiment when computer program is executed by processor
The step of survey method.
Each embodiment is described by the way of progressive in specification, the highlights of each of the examples are with other realities
Apply the difference of example, just to refer each other for identical similar portion between each embodiment.For device disclosed in embodiment
Speech, since it is corresponded to the methods disclosed in the examples, so description is fairly simple, related part is referring to method part illustration
.
Professional further appreciates that, with reference to each exemplary unit of the embodiments described herein description
And algorithm steps, can be realized with the combination of electronic hardware, computer software or the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is performed actually with hardware or software mode, specific application and design constraint depending on technical solution.Profession
Technical staff can realize described function to each specific application using distinct methods, but this realization should not
Think beyond the scope of this invention.
It can directly be held with reference to the step of method or algorithm that the embodiments described herein describes with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Above to a kind of nodule detection methods based on convolutional neural networks provided by the present invention, system, equipment and meter
Calculation machine readable storage medium storing program for executing is described in detail.Specific case used herein to the principle of the present invention and embodiment into
Elaboration is gone, the explanation of above example is only intended to facilitate the understanding of the method and its core concept of the invention.It should be pointed out that pair
For those skilled in the art, without departing from the principle of the present invention, the present invention can also be carried out
Some improvements and modifications, these improvement and modification are also fallen within the protection scope of the claims of the present invention.
Claims (11)
1. a kind of nodule detection methods based on convolutional neural networks, which is characterized in that the method includes：
The three-dimensional data of the tubercle of acquisition is analyzed using convolutional neural networks, obtain the classification information of the tubercle with
And location information；
Export the classification information and the location information.
2. according to the method described in claim 1, it is characterized in that, it is described using convolutional neural networks to the three of the tubercle of acquisition
Dimension volume data is analyzed, and obtains the classification information and location information of the tubercle, including：
The characteristic information of the network extraction three-dimensional data is extracted by three-dimensional feature；
Intensive cuboid frame sequence is generated according to the corresponding volume data of the three-dimensional data；
Using classifying, subnet handles the intensive cuboid frame sequence and the characteristic information, obtains the tubercle
Classification information；And/or
The intensive cuboid frame sequence and the characteristic information are handled using subnet is returned, obtain the tubercle
Cuboid encirclement frame；And/or
The cuboid encirclement frame is handled using subnet is divided, obtains the mask information of the tubercle.
It is 3. according to the method described in claim 2, it is characterized in that, described according to the corresponding volume data of the three-dimensional data
Intensive cuboid frame sequence is generated, including：
The three-dimensional data is detected using RPN networks, obtains candidate VOI regions；
Intensive cuboid frame sequence is generated according to the volume data of the corresponding three-dimensional data in the candidate VOI regions.
4. according to the method described in claim 3, it is characterized in that, it is described using divide subnet to the cuboid encirclement frame into
Row processing, obtains the mask information of the tubercle, including：
Non-maxima suppression calculating is carried out to cuboid encirclement frame, obtains accurate encirclement frame；
The volume data in the accurate encirclement frame is split using subnet is divided, obtains the mask information of the tubercle.
5. according to the method described in claim 1, it is characterized in that, the acquisition of the three-dimensional data, including：
The three-dimensional data is obtained using three dimensional ultrasound probe.
6. according to the method described in claim 5, it is characterized in that, it is described using convolutional neural networks to the three of the tubercle of acquisition
Dimension volume data is analyzed, and before obtaining classification information and the location information of the tubercle, is further included：
Resampling is carried out to the three-dimensional data of the tubercle of acquisition；
By the three-dimensional data after each resampling divided by the maximum value in the three-dimensional data of acquisition, and 0.5 is subtracted, returned
One changes volume data.
7. according to claim 1-6 any one of them methods, which is characterized in that the output classification information and described
Location information, including：
Three-dimensional rendering processing is carried out to the classification information, the location information and the pretreated volume data, visually
Change output treated data.
8. a kind of nodule detection system based on convolutional neural networks, which is characterized in that the system comprises：
Deep learning module for being analyzed using convolutional neural networks the three-dimensional data of the tubercle of acquisition, obtains institute
State the classification information and location information of tubercle；
Output module, for exporting the classification information and the location information.
9. system according to claim 8, which is characterized in that the output module is specially to visualize to export the classification
The module of information and the location information.
10. a kind of nodule detection equipment based on convolutional neural networks, which is characterized in that including：
Processor for being analyzed using convolutional neural networks the three-dimensional data of the tubercle of acquisition, obtains the tubercle
Classification information and location information；
Output device, for exporting the classification information and the location information.
11. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program is realized when the computer program is executed by processor and convolutional neural networks is based on as described in any one of claim 1 to 7
Nodule detection methods the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711464206.XA CN108171694B (en) | 2017-12-28 | 2017-12-28 | Method, system and equipment for detecting nodule based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711464206.XA CN108171694B (en) | 2017-12-28 | 2017-12-28 | Method, system and equipment for detecting nodule based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108171694A true CN108171694A (en) | 2018-06-15 |
| CN108171694B CN108171694B (en) | 2021-05-14 |
Family
ID=62519618
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711464206.XA Active CN108171694B (en) | 2017-12-28 | 2017-12-28 | Method, system and equipment for detecting nodule based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108171694B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108742679A (en) * | 2018-06-29 | 2018-11-06 | 上海联影医疗科技有限公司 | Nodule detection device and method |
| CN109146947A (en) * | 2018-09-04 | 2019-01-04 | 清华-伯克利深圳学院筹备办公室 | Marine fishes acquiring three-dimensional images and processing method, device, equipment and medium |
| CN109325943A (en) * | 2018-09-10 | 2019-02-12 | 深圳开立生物医疗科技股份有限公司 | A kind of three-D volumes measurement method and device |
| CN109815843A (en) * | 2018-12-29 | 2019-05-28 | 深圳云天励飞技术有限公司 | Object detection method and Related product |
| WO2020106216A1 (en) * | 2018-11-19 | 2020-05-28 | Medo Dx Pte. Ltd. | Method and system for generating a three-dimensional ultrasound image of a tissue volume |
| US10993653B1 (en) | 2018-07-13 | 2021-05-04 | Johnson Thomas | Machine learning based non-invasive diagnosis of thyroid disease |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740875A (en) * | 2016-03-17 | 2016-07-06 | 电子科技大学 | Pulmonary nodule multi-round classification method based on multi-scale three-dimensional block feature extraction |
| US9589374B1 (en) * | 2016-08-01 | 2017-03-07 | 12 Sigma Technologies | Computer-aided diagnosis system for medical images using deep convolutional neural networks |
| CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
| CN106780460A (en) * | 2016-12-13 | 2017-05-31 | 杭州健培科技有限公司 | A kind of Lung neoplasm automatic checkout system for chest CT image |
| CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
| CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
| CN107103187A (en) * | 2017-04-10 | 2017-08-29 | 四川省肿瘤医院 | The method and system of Lung neoplasm detection classification and management based on deep learning |
| CN107154043A (en) * | 2017-06-05 | 2017-09-12 | 杭州健培科技有限公司 | A kind of Lung neoplasm false positive sample suppressing method based on 3DCNN |
| CN107274402A (en) * | 2017-06-27 | 2017-10-20 | 北京深睿博联科技有限责任公司 | A kind of Lung neoplasm automatic testing method and system based on chest CT image |
| CN107369154A (en) * | 2017-07-19 | 2017-11-21 | 电子科技大学 | The detection method and device of image |
-
2017
- 2017-12-28 CN CN201711464206.XA patent/CN108171694B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740875A (en) * | 2016-03-17 | 2016-07-06 | 电子科技大学 | Pulmonary nodule multi-round classification method based on multi-scale three-dimensional block feature extraction |
| US9589374B1 (en) * | 2016-08-01 | 2017-03-07 | 12 Sigma Technologies | Computer-aided diagnosis system for medical images using deep convolutional neural networks |
| CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
| CN106780460A (en) * | 2016-12-13 | 2017-05-31 | 杭州健培科技有限公司 | A kind of Lung neoplasm automatic checkout system for chest CT image |
| CN107016665A (en) * | 2017-02-16 | 2017-08-04 | 浙江大学 | A kind of CT pulmonary nodule detection methods based on depth convolutional neural networks |
| CN106940816A (en) * | 2017-03-22 | 2017-07-11 | 杭州健培科技有限公司 | Connect the CT image Lung neoplasm detecting systems of convolutional neural networks entirely based on 3D |
| CN107103187A (en) * | 2017-04-10 | 2017-08-29 | 四川省肿瘤医院 | The method and system of Lung neoplasm detection classification and management based on deep learning |
| CN107154043A (en) * | 2017-06-05 | 2017-09-12 | 杭州健培科技有限公司 | A kind of Lung neoplasm false positive sample suppressing method based on 3DCNN |
| CN107274402A (en) * | 2017-06-27 | 2017-10-20 | 北京深睿博联科技有限责任公司 | A kind of Lung neoplasm automatic testing method and system based on chest CT image |
| CN107369154A (en) * | 2017-07-19 | 2017-11-21 | 电子科技大学 | The detection method and device of image |
Non-Patent Citations (4)
| Title |
|---|
| FANGZHOU LIAO等: "Evaluate the Malignancy of Pulmonary Nodules Using the 3D Deep Leaky Noisy-or Network", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
| QI DOU等: "Multi-level contextual 3d cnns for false positive reduction in pulmonary nodule detection", 《IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING》 * |
| 李云鹏等: "基于三维DBSCAN的疑似肺结节提取方法", 《现代盐化工》 * |
| 陈庆麟: "肺结节计算机辅助诊断系统关键技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108742679A (en) * | 2018-06-29 | 2018-11-06 | 上海联影医疗科技有限公司 | Nodule detection device and method |
| US10993653B1 (en) | 2018-07-13 | 2021-05-04 | Johnson Thomas | Machine learning based non-invasive diagnosis of thyroid disease |
| CN109146947A (en) * | 2018-09-04 | 2019-01-04 | 清华-伯克利深圳学院筹备办公室 | Marine fishes acquiring three-dimensional images and processing method, device, equipment and medium |
| CN109325943A (en) * | 2018-09-10 | 2019-02-12 | 深圳开立生物医疗科技股份有限公司 | A kind of three-D volumes measurement method and device |
| WO2020106216A1 (en) * | 2018-11-19 | 2020-05-28 | Medo Dx Pte. Ltd. | Method and system for generating a three-dimensional ultrasound image of a tissue volume |
| US11779309B2 (en) | 2018-11-19 | 2023-10-10 | Exo Imaging, Inc. | Method and system for generating a three-dimensional ultrasound image of a tissue volume from two-dimensional ultrasound images |
| CN109815843A (en) * | 2018-12-29 | 2019-05-28 | 深圳云天励飞技术有限公司 | Object detection method and Related product |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108171694B (en) | 2021-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108171694A (en) | Nodule detection methods, system and equipment based on convolutional neural networks | |
| Sevastopolsky | Optic disc and cup segmentation methods for glaucoma detection with modification of U-Net convolutional neural network | |
| Xu et al. | DeepLN: a framework for automatic lung nodule detection using multi-resolution CT screening images | |
| CN111476774B (en) | Intelligent sign recognition device based on novel coronavirus pneumonia CT detection | |
| CN111325739A (en) | Method and device for detecting lung focus and training method of image detection model | |
| CN110197492A (en) | A kind of cardiac MRI left ventricle dividing method and system | |
| US20220383661A1 (en) | Method and device for retinal image recognition, electronic equipment, and storage medium | |
| Choromanska et al. | Automatic reconstruction of neural morphologies with multi-scale tracking | |
| Palma et al. | Detection of masses and architectural distortions in digital breast tomosynthesis images using fuzzy and a contrario approaches | |
| CN110930414A (en) | Lung region shadow marking method and device of medical image, server and storage medium | |
| Cordeiro et al. | Segmentation of mammography by applying growcut for mass detection | |
| CN109063720A (en) | Handwritten word training sample acquisition methods, device, computer equipment and storage medium | |
| TWI587844B (en) | Medical image processing apparatus and breast image processing method thereof | |
| Hu et al. | Deep learning-based image evaluation for cervical precancer screening with a smartphone targeting low resource settings–Engineering approach | |
| Roy et al. | Automatic diagnosis of melanoma from dermoscopic image using real-time object detection | |
| CN109087310A (en) | Dividing method, system, storage medium and the intelligent terminal of Meibomian gland texture region | |
| CN108171696A (en) | A kind of placenta detection method and device | |
| Jiang et al. | Improving the generalizability of infantile cataracts detection via deep learning-based lens partition strategy and multicenter datasets | |
| CN111178420A (en) | Coronary segment labeling method and system on two-dimensional contrast image | |
| Nie et al. | Recent advances in diagnosis of skin lesions using dermoscopic images based on deep learning | |
| CN110021019A (en) | A kind of thickness distributional analysis method of the AI auxiliary hair of AGA clinical image | |
| US20230419492A1 (en) | Method and apparatus for providing information associated with immune phenotypes for pathology slide image | |
| Bernal et al. | A region segmentation method for colonoscopy images using a model of polyp appearance | |
| Xue et al. | Cracked tongue recognition based on deep features and multiple-instance SVM | |
| CN116778559A (en) | Face wrinkle three-dimensional evaluation method and system based on Gaussian process and random transformation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 518000 201, 202, building 12, Shenzhen Software Park (phase 2), No.1, Keji Middle Road, Maling community, Yuehai street, Nanshan District, Shenzhen City, Guangdong Province Applicant after: SONOSCAPE MEDICAL Corp. Address before: 518051 Guangdong city of Shenzhen province Nanshan District Yuquanlu Road Yizhe building 4, 5, 8, 9, 10 floor Applicant before: SONOSCAPE MEDICAL Corp. |
|
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200529 Address after: 430000 2 / F, building B13, biological industry (Jiufeng) innovation enterprise base, No. 666, Gaoxin Avenue, Donghu New Technology Development Zone, Wuhan City, Hubei Province Applicant after: Opening of biomedical technology (Wuhan) Co.,Ltd. Address before: 518000 201, 202, building 12, Shenzhen Software Park (phase 2), No.1, Keji Middle Road, Maling community, Yuehai street, Nanshan District, Shenzhen City, Guangdong Province Applicant before: SONOSCAPE MEDICAL Corp. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |