CN110335668A - Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning - Google Patents
Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning Download PDFInfo
- Publication number
- CN110335668A CN110335668A CN201910428177.4A CN201910428177A CN110335668A CN 110335668 A CN110335668 A CN 110335668A CN 201910428177 A CN201910428177 A CN 201910428177A CN 110335668 A CN110335668 A CN 110335668A
- Authority
- CN
- China
- Prior art keywords
- pathology
- model
- training
- image
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Public Health (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Library & Information Science (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning, which comprises the following steps: s1, constructing a cloud platform for digital pathological labeling to obtain a labeled training set, a labeled verification set and a labeled test set; s2, creating a database for storing pathological digital full-section images; s3, training a convolutional neural network model by using the training set, the verification set and the test set to obtain a digital pathological model; and S4, calling the digital pathology model to perform digital pathology image analysis on the pathology digital full-section image. The cloud platform for data annotation is provided, data transmission and annotation are carried out on line, no special equipment requirement exists, and the efficiency of annotation and data transmission is improved; and constructing a database for storing the full-slice images, and acquiring corresponding pathology digital full-slice images from the database in the labeling process, the training process and the subsequent analysis and detection process, so as to realize the analysis of the full-slice digital pathology with very large auxiliary data flow.
Description
Technical field
The invention belongs to pathology figure assistant analysis technical fields, more particularly, to a kind of thyroid gland based on deep learning
Cancer cell pathology figure aided analysis method and system.
Background technique
Pathological diagnosis is needed to finally making a definite diagnosis for the lesions such as thyroid nodule, Fine-needle Aspiration Tissuess (FNA) are economical, operate
Simplicity, wound is small, accuracy rate is higher, and the accuracy of lesion diagnosis can be improved.But due to the tissue mass of fine needle puncture and its
Cell component is less, tissue morphology in sample and cytoplasm structure are largely lost, and with puncture and cytodiagnosis
Experience is related, and tubercle is too small, needle carefully all may result in very much and draw materials insufficient or cannot accurately draw materials in diseased region.By institute
Take tissue few, film-making, diagosis are more difficult etc. can influence inspection result, be generally difficult to determine lesion type, give hospital and doctor
It makes Accurate Diagnosis and brings very big difficulty, since thyroid cancer 90% is the above are papillary carcinoma, critical issue is nipple
The diagnosis of shape cancer.
In recent years, the digitlization process of medical field is constantly promoted to longitudinal direction, and is gradually stepped to the intelligent medical treatment stage
Into.With the high speed development and maturation of artificial intelligence (Artificial Intelligence, AI) technology, in image recognition, depth
The breakthrough of the key technology areas such as study, neural network is spent, artificial intelligence+medical treatment intelligent mode has turned on, is mainly used in
The various aspects such as data acquisition, auxiliary diagnosis, detection feedback, teaching, training, accurate medical treatment.And in artificial intelligence technology, figure
Picture identification is the field that it breaks through at first, has more than 90% in medical data at present from medical imaging image, these data are big
It mostly to carry out manual analysis and compare, if it is possible to use algorithm automated analysis image, then by image and other case notes
It compares, clinical diagnosis efficiency can be greatly improved and reduces medical misdiagnosis rate, accurate diagnosis is made in help.
In order to realize above-mentioned imagination, people have carried out long-term exploration, such as Chinese patent discloses a kind of digestive tract disease
Stove image identification system and recognition methods [application number: CN201610405322.3], step 1: being stored in engineering in memory
Practise training data, wherein the machine learning training data includes training sample image, test sample image, image classification letter
Breath, data resolution module extraction machine learning training data from memory, and the machine learning training data of extraction is carried out
Format conversion, the image of picture format needed for generating；
Step 2: image characteristics extraction module uses scale invariant feature transfer algorithm and complete local binary patterns algorithm
The image texture characteristic of training sample image in extraction machine learning training data, while using super-pixel method and gridding method pair
Machine learning training data is split, and then extracts the entropy feature of training sample image in machine learning training data after dividing
With color moment characteristics；Image texture characteristic, entropy feature and color moment characteristics are also transferred to machine by image characteristics extraction module respectively
Device study module and picture recognition module；Machine learning training data is also transferred to machine learning mould by image characteristics extraction module
Block；
Step 3: machine learning module is using the deep learning method of convolutional neural networks according to image classification information to machine
Device learning training data carry out the classification of alimentary canal position, obtain alimentary canal position classification data, machine learning module is also according to figure
As textural characteristics, entropy feature and color moment characteristics, and learning training is carried out according to algorithm of support vector machine and generates alimentary canal lesion
Information Data model；The classification of alimentary canal position and alimentary canal lesion Information Data model are also transferred to image by machine learning module
Identification module；
Step 4: picture recognition module classifies to machine learning training data according to alimentary canal position classification data, and
Result based on the classification of alimentary canal position extracts image texture characteristic, entropy to the machine learning training data at different alimentary canal positions
Then feature and color moment characteristics are extracted using machine learning training data of the Adaboost algorithm to different alimentary canal positions
Image texture characteristic, entropy feature and color moment characteristics carry out lesion and identify to obtain the suspicious region of lesion, finally using support to
Amount machine application alimentary canal lesion Information Data model is classified to obtain accurate lesion information to lesion identification suspicious region.
Above scheme application artificial intelligence technology, to being promoted, curative activity efficiency, the accuracy for improving doctor's judgement etc. are square
Face plays good booster action.But above scheme can only realize the recognition efficiency and accuracy for improving image, it can not
The analysis of the very big full slice number pathology of auxiliary data amount, also because the reasons such as data volume is big can not be applied to cell pathology
With the full slice image analysis of histopathology.
Summary of the invention
Regarding the issue above, the present invention provides a kind of thyroid carcinoma cell pathology figure based on deep learning is auxiliary
Help analysis method；
It is another object of the present invention in view of the above-mentioned problems, providing a kind of disease based on deep learning using the above method
Manage visual aids analysis system.
In order to achieve the above objectives, present invention employs following technical proposals:
A kind of thyroid carcinoma cell pathology figure aided analysis method based on deep learning, comprising the following steps:
S1. building obtains the training set by mark, verifying collection and test set for the cloud platform of digital pathology mark；
S2. the database for storing pathology number full slice image is created；
S3. convolutional neural networks model is trained to obtain number using the training set, verifying collection and test set
Pathological model；
S4. call number pathological model is to carry out digital pathological image analysis to pathology number full slice image.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, the pathology number is complete
Sectioning image includes the full slice image for the full slice image of cell pathology and for histopathology；
And in step s3, the digital pathological model includes for the cell pathology model of cell pathology and for group
Knit the histopathology model of pathology.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, in step s3, group
Knitting pathological model, training obtains by the following method:
S301. model training is carried out to ScanNet network model using histopathology training set；
S302. model checking is carried out to the model after training using histopathology verifying collection, executes step if verification passes through
Otherwise rapid S303 adjusts network structure or parameter and return step S301；
S303. trained model is tested using histopathology test set with verify model whether there is it is quasi-
Conjunction problem, and if it exists, then adjust network structure or parameter and return step S301.
It is in step s3, right in the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning
The training of cell pathology model includes checking screening system for the training of the preliminary screening system of stage one and for the stage two
Training.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, the preliminary screening of stage one
The training process of system includes:
S3121. model training is carried out to ScanNet network model using cell pathology training set；
S3122. model checking is carried out to the model after training using cell pathology verifying collection, executes step if verification passes through
Otherwise rapid S3123 adjusts network structure or parameter and return step S3121；
S3123. the ScanNet model by step S3122 processing is tested to test using cell pathology test set
Model of a syndrome whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3121.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, the stage two checks screening
The training process of system includes:
S3131. two training set of stage, the verifying of stage two collection and two test set of stage of cell pathology are constructed；
S3132. two training set of service stage training depth residual error network model；
S3133. the verifying of service stage two collection verifies the model after training, if verification is by thening follow the steps
Otherwise S3134 adjusts network structure or parameter and return step S3132；
S3134. two test set of service stage to by step S3133 processing depth residual error network model tested with
Verifying model whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3132.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, cell pathology mould is used
Type carry out Cellular Pathology Image analysis process include:
S41. picture pretreatment is carried out to pathology number full slice image；
S42. one screening of stage is carried out to obtain to pretreated full slice image using ScanNet preliminary screening system
Candidate target；
S313. the review screening of stage two is carried out to obtain prediction knot to candidate target using DresNet review screening system
Fruit.
In the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning, pass through successive ignition mistake
The corresponding network model of Cheng Xunlian, and in each iterative process excavate and add the classification results of mistake to training sample with
Network model is set to improve the training proportion of difficult case.
It is raw using virtual image in the above-mentioned thyroid carcinoma cell pathology figure aided analysis method based on deep learning
At technology, merges emulation by existing image resource random groups and generate virtual pathology image to expand training sample.
A kind of thyroid cancer of the thyroid carcinoma cell pathology figure aided analysis method based on described based on deep learning is thin
Born of the same parents' pathological image Computer Aided Analysis System, including cloud platform, database, deep learning model and decision system, wherein
Cloud platform, for from database receive pathology number full slice image, and for medical practitioner in platform to disease
It manages digital full slice image and carries out digital pathology mark；
Database, for receiving and storing pathology number full slice image and training sample；
Deep learning model, for carrying out model training, verifying and test to convolutional neural networks using training sample, with
Obtain digital pathological model；
Decision system, for obtaining pathology number full slice image to be analyzed and call number pathology mould from database
Type is to carry out digital pathological image analysis to the pathology number full slice image.
The present invention has the advantages that 1 uses digitlization pathology system, has and be easy to save, transmit, managing and facilitating browsing
The advantages that；2, the cloud platform marked for data, data transmission and the online upper progress of mark are provided, and not special equipment needs
It asks, improves the efficiency of mark and data transmission；3, the database for storing full slice image is constructed, annotation process was trained
Journey and analyte detection process later obtain corresponding pathology number full slice image from database, to realize supplementary number
According to the analysis for flowing very big full slice number pathology；4, under basic framework, residual error network structure model is introduced, improves network
Resolution capability of the model to lesion.
Detailed description of the invention
Fig. 1 is the method for thyroid carcinoma cell pathology figure aided analysis method of the embodiment of the present invention one based on deep learning
Flow chart；
Fig. 2 is the full slice image of cell pathology in the embodiment of the present invention one；
Fig. 3 is the enlarged drawing that position is irised out in Fig. 1；
The full slice image of histopathology in Fig. 4 embodiment of the present invention one；
Fig. 5 is the enlarged drawing that position is irised out in Fig. 3；
Fig. 6 is the training flow chart of one thyroid carcinoma cell histopathology model of the embodiment of the present invention；
Fig. 7 is the training flow chart of the preliminary screening system of one stage of the embodiment of the present invention one；
Fig. 8 is the training flow chart of one stage two of embodiment of the present invention review screening system；
Fig. 9 is the method flow diagram that the embodiment of the present invention one carries out cell pathology analysis using cell pathology model；
Figure 10 is the structural block diagram of two thyropathy of embodiment of the present invention reason visual aids analysis system.
Appended drawing reference: cloud platform 1, database 2, deep learning model 3, decision system 4.
Specific embodiment
The present invention will be further described in detail with reference to the accompanying drawings and detailed description.
Embodiment one
As shown in Figure 1, present embodiment discloses a kind of thyroid carcinoma cell pathology figure assistant analysis based on deep learning
Method, mainly for the pathological image assistant analysis of papillary thyroid carcinoma cells, specifically includes the following steps:
S1. building obtains the training set by mark, verifying collection and test set for the cloud platform of digital pathology mark；
S2. the database for storing pathology number full slice image is created；
S3. convolutional neural networks model is trained to obtain number using the training set, verifying collection and test set
Pathological model；
S4. the digital pathological image point of pathology number full slice image is obtained by decision system call number pathological model
Analyse result.
Different from traditional images, full slice data are very big, and in general situation, the size of normal picture is all in 10K-
The magnitude of 10M.But full slice image (WSI) picture generally occupies the space of 1G-10G.The present embodiment constructs dedicated for depositing
The database of pathology number full slice image is stored up, and disposes full slice scanner in hospital, full slice scanner is direct or indirect
Interworking Data library by database batch, automatically picks up and arranges digital pathological data, and reasonably managed.By data
Library is divided into historical data reservoir area, sample data reservoir area and database community to be analyzed, as sample pathology number full slice image,
History pathology number full slice image and pathology number full slice image to be analyzed are respectively stored in sample data reservoir area, history
Database community and database community to be analyzed, in order to cloud platform or decision system is extracted and the retraining in later period.
In addition, the present embodiment is by building one cloud platform for digital pathology mark, it can be artificial intelligent training
The preparation of data provides very big help.
It should be noted that depth learning technology is the technology of current comparative maturity, therefore here only to the weight of the present embodiment
Point process is illustrated, and is no longer described in detail to the specific training process of convolutional neural networks.In addition, when coming into operation, entirely
Slice scanner obtains pathology number full slice image, huge due to scanning the pathology number full slice amount of images got every time
Greatly, so decision system receives the pathology number full slice image that full slice scanner Current Scan arrives, and call number in batches
Output analysis is as a result, process of calculation analysis and ability after pathological model carries out calculation processing to each pathology number full slice image
Field technique personnel choose network model framework, algorithm and model training process is related, specific process of calculation analysis not herein into
Row repeats.
The present embodiment focuses on by constructing for the cloud platform of digital pathology mark and for storing full slice figure
The database of picture is realized the diagnosis of the very big full slice number pathology of auxiliary data amount based on depth learning technology, borrowed simultaneously
The cloud platform for digital pathology mark is helped, so that data is marked and is transmitted can carry out on line, not have special installation demand, greatly
The big efficiency for improving mark, data transmission and arranging.
Specifically, pathology number full slice image includes for the full slice image of cell pathology and for histopathology
Full slice image, wherein the full slice image of cell pathology is as shown in Figures 2 and 3, full slice image such as Fig. 4 of histopathology and
Shown in Fig. 5.
And in step s3, digital pathological model includes for the cell pathology model of cell pathology and for histopathology
Histopathology model.
So each area of database is respectively divided into corresponding cell pathology data field and histopathology data field, example here
If historical data area is divided into cell pathology historical data area and histopathology historical data area, sample data area is divided into cell
Pathology sample data area and resistance value pathology sample data area.Explanation is labeled by taking cell pathology as an example below: by medical practitioner
It is operated in cloud platform, successively transfers the cell pathology number full slice image in cell pathology historical data reservoir area in cloud
It is labeled in platform, the cell pathology number full slice image being each marked is as training sample and is put into cell pathology sample
The amount of database, mark is voluntarily determined that mark is more by mark personnel, is more conducive to the model training in later period, in addition, by marking
Other regions are deleted or moved to cell pathology number full slice image after note from cell pathology historical data area.Then, exist
In cloud platform, cell training set, cell verifying collection and cell tests collection three parts will be divided by the training sample of mark with right
Convolutional neural networks model is trained to obtain corresponding cell pathology model, and training sample is with training set, verifying collection and surveys
Examination collection=2:1:1 mode is distributed.
Specifically, as shown in fig. 6, in step s3, training obtains histopathology model by the following method:
S301. picture pretreatment is carried out to all training samples of histopathology；
S302. model training is carried out to ScanNet network model using training set；
S303. model checking is carried out to the model after training using histopathology verifying collection, executes step if verification passes through
Otherwise rapid S304 adjusts network structure or parameter and return step S302；
S304. trained model is tested using histopathology test set with verify model whether there is it is quasi-
Conjunction problem, and if it exists, then adjust network structure or parameter and return step S302, then obtain preferably by training if it does not exist
Histopathology model.
When coming into operation, when needing to analyze histopathology number full slice image, histopathology number is obtained
Word full slice image simultaneously calls histopathology model to carry out calculating the i.e. exportable final analysis/test knot of analysis to full slice image
Fruit.
Preferably, the cell typing of cell pathology is more, so including here being directed to rank to the training of cell pathology model
The training of one preliminary screening system of section and the training that screening system is checked for the stage two.
Specifically, as shown in fig. 7, the training process of the preliminary screening system of stage one includes:
S3121. picture pretreatment is carried out to the training sample of cell pathology；
S3122. model training is carried out to ScanNet network model using training set；
S3123. model checking is carried out to the model after training using cell pathology verifying collection, executes step if verification passes through
Otherwise rapid S3124 adjusts network structure or parameter and return step S3122；
S3124. the ScanNet model by step S3123 processing is tested to test using cell pathology test set
Model of a syndrome whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3122.
As shown in figure 8, the training process that the stage two checks screening system includes:
S3131. the training sample of cell pathology is constructed, and picture pretreatment is carried out to training sample, training sample is divided into
Two training set of stage, the verifying of stage two collection and two test set of stage；
S3132. two training set of service stage training depth residual error network model；
S3133. the verifying of service stage two collection verifies the model after training, if verification is by thening follow the steps
Otherwise S3134 adjusts network structure or parameter and return step S3132；
S3134. two test set of service stage to by step S3133 processing depth residual error network model tested with
Verifying model whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3132.
Specifically, as shown in figure 9, including: using the process that cell pathology model carries out cell pathology analysis
S41. picture pretreatment is carried out to pathology number full slice image；
S42. one screening of stage is carried out to obtain to pretreated full slice image using ScanNet preliminary screening system
Candidate target；
S313. the review screening of stage two is carried out to obtain prediction knot to candidate target using DresNet review screening system
Fruit.
The present embodiment is based on current ScanNet basic framework for cell pathology model, introduces the minds such as residual error network structure
Through network model, cell pathology model is improved to the resolution capability of lesion.
Preferably, it here by the corresponding network model of successive ignition process training, and is dug in each iterative process
Pick and the wrong classification results of addition are to training sample so that network model improves the training proportion of difficult case.
In addition, the acquisition of pathological marker data is and its expensive, the present embodiment uses virtual image generation technique, such as
Confrontation generates network (GAN), by existing image resource, random combine, and emulates and generates virtual pathology image, and then expand
Our training sample.For one side, this will greatly reduce procurement cost to medical image；On the other hand,
This is beneficial to the generalization ability for improving neural network.
The pathological data collection scale of construction is huge, however significant portion of example is relatively easy to distinguish, if to all examples
Learning training is alike carried out, entire training process can be made to become inefficient, so the present embodiment is to different training samples
Different study weights is given, allows network that can focus more on difficult example in the training process, improves the training effect of network
Rate, to improve the resolution capability of model.
Further, for each digital pathological model, the present embodiment preferably trains at least two network models simultaneously, and
Actively reduce the weight that the high noise sample of inconsistency is generally acknowledged between multiple network models.It is well known that pathological marker is one
It is strongly dependent on a professional work of doctors experience.Due to the difference of the working experience level of different doctors, this is also inevitable
Resulting in the sample of label, there are certain disagreements, thus can not 100% accuracy for guaranteeing all training datas labels, this reality
Apply example and pass through the weight for actively reducing the noise sample that generally acknowledged inconsistency is high between multiple network models, with guarantee training effect from
And improve the overall performance that final digital pathological model identifies lesion.
Embodiment two
As shown in Figure 10, present embodiment discloses a kind of according to embodiment one based on the thyroid cancer of deep learning
The pathological image Computer Aided Analysis System of cell pathology figure aided analysis method, including cloud platform 1, database 2, deep learning model
3 and decision system 4, wherein
Cloud platform 1, for from database receive pathology number full slice image, and for medical practitioner in platform to disease
It manages the digital digital pathology of full slice image progress to mark, to obtain the training sample by marking, i.e. training set, verifying collects and survey
Examination collection；
Database 2, for receiving and storing pathology number full slice image and by the training sample of mark, training sample
Including training set, verifying collection and test set；
Deep learning model 3, for carrying out model to convolutional neural networks respectively using training set, verifying collection and test set
Training, verifying and test, to obtain digital pathological model；
Decision system 4, for obtaining pathology number full slice image to be analyzed and call number pathology from database
Model is to carry out digital pathological image analysis to the pathology number full slice image.
Wherein decision system 4 and cloud platform 1 can be accessed by terminals such as computers, knot is marked and analyzed for pathology
The acquisition of fruit.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Although the terms such as cloud platform 1, database 2, deep learning model 3 and decision system 4 are used more herein,
But it does not exclude the possibility of using other terms.The use of these items is only for be more convenient to describe and explain the present invention
Essence；Being construed as any additional limitation is disagreed with spirit of that invention.
Claims (10)
1. a kind of thyroid carcinoma cell pathology figure aided analysis method based on deep learning, which is characterized in that including following step
It is rapid:
S1. building obtains the training set by mark, verifying collection and test set for the cloud platform of digital pathology mark；
S2. the database for storing pathology number full slice image is created；
S3. convolutional neural networks model is trained using the training set, verifying collection and test set to obtain digital pathology
Model；
S4. call number pathological model is to carry out digital pathological image analysis to pathology number full slice image.
2. the thyroid carcinoma cell pathology figure aided analysis method according to claim 1 based on deep learning, feature
It is, the pathology number full slice image includes the full slice for the full slice image of cell pathology and for histopathology
Image；
And in step s3, the digital pathological model includes for the cell pathology model of cell pathology and for tissue disease
The histopathology model of reason.
3. the thyroid carcinoma cell pathology figure aided analysis method according to claim 2 based on deep learning, feature
It is, in step s3, training obtains histopathology model by the following method:
S301. model training is carried out to ScanNet network model using histopathology training set；
S302. model checking is carried out to the model after training using histopathology verifying collection, if verification is by thening follow the steps
Otherwise S303 adjusts network structure or parameter and return step S301；
S303. trained model is tested using histopathology test set to verify model with the presence or absence of over-fitting and ask
Topic, and if it exists, then adjust network structure or parameter and return step S301.
4. the thyroid carcinoma cell pathology figure aided analysis method according to claim 3 based on deep learning, feature
It is, in step s3, the training to cell pathology model includes for the training of the preliminary screening system of stage one and for rank
Section two checks the training of screening system.
5. the thyroid carcinoma cell pathology figure aided analysis method according to claim 4 based on deep learning, feature
It is, the training process of preliminary screening system of stage one includes:
S3121. model training is carried out to ScanNet network model using cell pathology training set；
S3122. model checking is carried out to the model after training using cell pathology verifying collection, if verification is by thening follow the steps
Otherwise S3123 adjusts network structure or parameter and return step S3121；
S3123. the ScanNet model by step S3122 processing is tested to verify mould using cell pathology test set
Type whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3121.
6. the thyroid carcinoma cell pathology figure aided analysis method according to claim 5 based on deep learning, feature
It is, the training process that the stage two checks screening system includes:
S3131. two training set of stage, the verifying of stage two collection and two test set of stage of cell pathology are constructed；
S3132. two training set of service stage training depth residual error network model；
S3133. the verifying of service stage two collection verifies the model after training, no if verification is by thening follow the steps S3134
Then adjust network structure or parameter and return step S3132；
S3134. two test set of service stage tests to verify the depth residual error network model by step S3133 processing
Model whether there is overfitting problem, and if it exists, then adjust network structure or parameter and return step S3132.
7. the thyroid carcinoma cell pathology figure aided analysis method according to claim 6 based on deep learning, feature
It is, includes: using the process that cell pathology model carries out Cellular Pathology Image analysis
S41. picture pretreatment is carried out to pathology number full slice image；
S42. one screening of stage is carried out to obtain candidate to pretreated full slice image using ScanNet preliminary screening system
Target；
S313. the review screening of stage two is carried out to obtain prediction result to candidate target using DresNet review screening system.
8. the thyroid carcinoma cell pathology figure assistant analysis side described in -6 any one based on deep learning according to claim 1
Method, which is characterized in that by the corresponding network model of successive ignition process training, and excavate and add in each iterative process
Add the classification results of mistake to training sample so that network model improves the training proportion of difficult case.
9. the thyroid carcinoma cell pathology figure aided analysis method according to claim 8 based on deep learning, feature
Be, using virtual image generation technique, by existing image resource random groups merge emulation generate virtual pathology image with
Expand training sample.
10. the thyroid carcinoma cell pathology figure auxiliary point described in a kind of -9 any one according to claim 1 based on deep learning
The thyroid carcinoma cell pathological image Computer Aided Analysis System of analysis method, which is characterized in that including cloud platform (1), database (2),
Deep learning model (3) and decision system (4), wherein
Cloud platform (1), for from database receive pathology number full slice image, and for medical practitioner in platform to pathology
Digital full slice image carries out digital pathology mark；
Database (2), for receiving and storing pathology number full slice image and training sample；
Deep learning model (3), for carrying out model training, verifying and test to convolutional neural networks using training sample, with
Obtain digital pathological model；
Decision system (4), for obtaining pathology number full slice image to be analyzed and call number pathology mould from database
Type is to carry out digital pathological image analysis to the pathology number full slice image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910428177.4A CN110335668A (en) | 2019-05-22 | 2019-05-22 | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910428177.4A CN110335668A (en) | 2019-05-22 | 2019-05-22 | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110335668A true CN110335668A (en) | 2019-10-15 |
Family
ID=68139170
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910428177.4A Pending CN110335668A (en) | 2019-05-22 | 2019-05-22 | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110335668A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111134735A (en) * | 2019-12-19 | 2020-05-12 | 复旦大学附属中山医院 | Lung cell pathology rapid on-site evaluation system and method and computer readable storage medium |
| CN111460173A (en) * | 2019-12-26 | 2020-07-28 | 四川大学华西医院 | Method for constructing disease ontology model of thyroid cancer |
| CN111598875A (en) * | 2020-05-18 | 2020-08-28 | 北京小白世纪网络科技有限公司 | Method, system and device for building thyroid nodule automatic detection model |
| CN111598876A (en) * | 2020-05-18 | 2020-08-28 | 北京小白世纪网络科技有限公司 | Method, system and equipment for constructing thyroid nodule automatic identification model |
| CN111709941A (en) * | 2020-06-24 | 2020-09-25 | 上海迪影科技有限公司 | Lightweight automatic deep learning system and method for pathological image |
| CN113139930A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification method and device, computer equipment and storage medium |
| CN114898866A (en) * | 2022-05-24 | 2022-08-12 | 广州锟元方青医疗科技有限公司 | Thyroid cell auxiliary diagnosis method, equipment and storage medium |
| CN116580011A (en) * | 2023-05-19 | 2023-08-11 | 北京透彻未来科技有限公司 | Endometrial cancer full-slide image detection system of deep learning model |
| CN117373695A (en) * | 2023-10-12 | 2024-01-09 | 北京透彻未来科技有限公司 | Extreme deep convolutional neural network-based diagnosis system for diagnosis of cancer disease |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368670A (en) * | 2017-06-07 | 2017-11-21 | 万香波 | Stomach cancer pathology diagnostic support system and method based on big data deep learning |
| CN108364293A (en) * | 2018-04-10 | 2018-08-03 | 复旦大学附属肿瘤医院 | A kind of on-line training thyroid tumors Ultrasound Image Recognition Method and its device |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN108764329A (en) * | 2018-05-24 | 2018-11-06 | 复旦大学附属华山医院北院 | A kind of construction method of lung cancer pathology image data set |
| CN109300530A (en) * | 2018-08-08 | 2019-02-01 | 北京肿瘤医院 | The recognition methods of pathological picture and device |
-
2019
- 2019-05-22 CN CN201910428177.4A patent/CN110335668A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368670A (en) * | 2017-06-07 | 2017-11-21 | 万香波 | Stomach cancer pathology diagnostic support system and method based on big data deep learning |
| CN108364293A (en) * | 2018-04-10 | 2018-08-03 | 复旦大学附属肿瘤医院 | A kind of on-line training thyroid tumors Ultrasound Image Recognition Method and its device |
| CN108717554A (en) * | 2018-05-22 | 2018-10-30 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors histopathologic slide image classification method and its device |
| CN108764329A (en) * | 2018-05-24 | 2018-11-06 | 复旦大学附属华山医院北院 | A kind of construction method of lung cancer pathology image data set |
| CN109300530A (en) * | 2018-08-08 | 2019-02-01 | 北京肿瘤医院 | The recognition methods of pathological picture and device |
Non-Patent Citations (1)
| Title |
|---|
| 李佐君: "《大数据的架构技术与应用实践的研究》", 30 April 2019, 东北师范大学出版社 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111134735A (en) * | 2019-12-19 | 2020-05-12 | 复旦大学附属中山医院 | Lung cell pathology rapid on-site evaluation system and method and computer readable storage medium |
| CN111460173A (en) * | 2019-12-26 | 2020-07-28 | 四川大学华西医院 | Method for constructing disease ontology model of thyroid cancer |
| CN111460173B (en) * | 2019-12-26 | 2023-02-03 | 四川大学华西医院 | Method for constructing disease ontology model of thyroid cancer |
| CN111598876A (en) * | 2020-05-18 | 2020-08-28 | 北京小白世纪网络科技有限公司 | Method, system and equipment for constructing thyroid nodule automatic identification model |
| CN111598876B (en) * | 2020-05-18 | 2021-03-16 | 北京小白世纪网络科技有限公司 | Method, system and equipment for constructing thyroid nodule automatic identification model |
| CN111598875A (en) * | 2020-05-18 | 2020-08-28 | 北京小白世纪网络科技有限公司 | Method, system and device for building thyroid nodule automatic detection model |
| CN111709941A (en) * | 2020-06-24 | 2020-09-25 | 上海迪影科技有限公司 | Lightweight automatic deep learning system and method for pathological image |
| CN111709941B (en) * | 2020-06-24 | 2023-05-09 | 上海迪影科技有限公司 | Lightweight automatic deep learning system and method for pathological image |
| CN113139930A (en) * | 2021-03-17 | 2021-07-20 | 杭州迪英加科技有限公司 | Thyroid slice image classification method and device, computer equipment and storage medium |
| CN113139930B (en) * | 2021-03-17 | 2022-07-15 | 杭州迪英加科技有限公司 | Thyroid slice image classification method and device, computer equipment and storage medium |
| CN114898866A (en) * | 2022-05-24 | 2022-08-12 | 广州锟元方青医疗科技有限公司 | Thyroid cell auxiliary diagnosis method, equipment and storage medium |
| CN114898866B (en) * | 2022-05-24 | 2024-03-15 | 广州锟元方青医疗科技有限公司 | Thyroid cell auxiliary diagnosis method, thyroid cell auxiliary diagnosis equipment and thyroid cell auxiliary diagnosis storage medium |
| CN116580011A (en) * | 2023-05-19 | 2023-08-11 | 北京透彻未来科技有限公司 | Endometrial cancer full-slide image detection system of deep learning model |
| CN117373695A (en) * | 2023-10-12 | 2024-01-09 | 北京透彻未来科技有限公司 | Extreme deep convolutional neural network-based diagnosis system for diagnosis of cancer disease |
| CN117373695B (en) * | 2023-10-12 | 2024-05-17 | 北京透彻未来科技有限公司 | Extreme deep convolutional neural network-based diagnosis system for diagnosis of cancer disease |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110335668A (en) | Thyroid cancer cell pathological map auxiliary analysis method and system based on deep learning | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN110491502B (en) | Microscope video stream processing method, system, computer device and storage medium | |
| CN109919230B (en) | Medical image pulmonary nodule detection method based on cyclic feature pyramid | |
| CN113724228B (en) | Tongue color and tongue fur color identification method and device, computer equipment and storage medium | |
| CN107280697A (en) | Lung neoplasm grading determination method and system based on deep learning and data fusion | |
| CN110491480A (en) | A kind of medical image processing method, device, electromedical equipment and storage medium | |
| CN107680088A (en) | Method and apparatus for analyzing medical image | |
| CN108596046A (en) | A kind of cell detection method of counting and system based on deep learning | |
| CN107330263A (en) | A kind of method of area of computer aided breast invasive ductal carcinoma histological grading | |
| CN106202955A (en) | Diagnosis associated packets method and system based on intellectual coded adaptation | |
| CN112086197B (en) | Breast nodule detection method and system based on ultrasonic medicine | |
| CN108846828A (en) | A kind of pathological image target-region locating method and system based on deep learning | |
| CN108010013A (en) | A kind of lung CT image pulmonary nodule detection methods | |
| WO2021027152A1 (en) | Image synthesis method based on conditional generative adversarial network, and related device | |
| CN112820399A (en) | Method and device for automatically diagnosing benign and malignant thyroid nodules | |
| CN114399634B (en) | Three-dimensional image classification method, system, equipment and medium based on weak supervision learning | |
| CN110807754B (en) | Fungus microscopic image segmentation detection method and system based on deep semantic segmentation | |
| Vuppala et al. | Explainable deep learning methods for medical imaging applications | |
| CN117237351A (en) | Ultrasonic image analysis method and related device | |
| CN116759076A (en) | Unsupervised disease diagnosis method and system based on medical image | |
| CN117174238A (en) | Automatic pathology report generation method based on artificial intelligence | |
| Dang et al. | Kashin-Beck disease diagnosis based on deep learning from hand X-ray images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191015 |
|
| RJ01 | Rejection of invention patent application after publication |