CN108766555A - The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy - Google Patents
The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy Download PDFInfo
- Publication number
- CN108766555A CN108766555A CN201810306839.6A CN201810306839A CN108766555A CN 108766555 A CN108766555 A CN 108766555A CN 201810306839 A CN201810306839 A CN 201810306839A CN 108766555 A CN108766555 A CN 108766555A
- Authority
- CN
- China
- Prior art keywords
- neuroendocrine tumors
- grade malignancy
- pancreatic neuroendocrine
- data
- convolutional layer
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Medicines Containing Antibodies Or Antigens For Use As Internal Diagnostic Agents (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a kind of computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy, method includes:Obtain initial data;Initial data is divided as unit of patient, obtains training set, test set and verification collection;It is trained and tests using convolutional neural networks according to training set, test set and verification collection, obtain Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;Patient data to be diagnosed is inputted into Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, exports corresponding Pancreatic Neuroendocrine Tumors grade malignancy.The present invention realizes the computer diagnosis to Pancreatic Neuroendocrine Tumors grade malignancy by deep learning, keep the result of diagnostic imaging more objective and accurate, and the probability of mistaken diagnosis is reduced to a certain extent, while the diagosis burden of image department doctor can be reduced, improve the working efficiency of doctors.It the composite can be widely applied to medical computer assistance application field.
Description
Technical field
The present invention relates to medical computer assistance application field, especially a kind of Pancreatic Neuroendocrine Tumors grade malignancy
Computer diagnosis method and system.
Background technology
Name Resolution:
CT:Computed Tomography, i.e. computed tomography.
Pathologic finding:Pathomorphology method for checking the pathological change in biological organs, tissue or cell.To visit
The lysis that organ, tissue or cell are occurred is begged for, the method that certain Pathomorphology inspection can be used checks that they are sent out
Raw lesion inquires into lesion Producing reason, pathogenesis, the occurrence and development process of lesion, finally makes pathological diagnosis.Pathology
Morphologic inspection method, looks first at the pathological change of gross specimen, then cuts a certain size pathological tissues, uses pathology
Pathological section is made in Histological method, and lesion is further checked with microscope.
Living body puncture:Refer to coherent video (such as B ultrasound, x-ray) guiding or (and) under positioning scenarios with special
Puncture needle guided from body surface to iconography, the process that the lesion that positions or specific position are punctured, be chiefly used in lesion
Area takes sample to carry out pathological biopsy.
Enhancing scanning:One of CT scan technology is applied to the scanning of intravascular contrast medium.Contain iodine organification through being injected intravenously
Object (i.e. contrast agent) is closed, general 60% cardiografin 1.5~2.0ml/kg rapid intravenous injections make amount of iodine in blood maintain one
Determine level, organ and lesion Imaging enhanced and show and become apparent from, is mainly used for differentiating that lesion is vascular or non-vascular, define
Longitudinal disease becomes the relationship with cardiovascular injuries, understands the blood supply situation of lesion to help to differentiate benign and malignant lesions etc., increases disease
The information content of stove is even clarified a diagnosis in order to carry out qualitative analysis to lesion.
Three-phase scan:Using the fireballing advantage of CT scan, after once intravenous injection contrast medium, according to detector
The blood supply feature of official, strengthening different periods respectively carries out twice or repeatedly complete helical scanning the organ of inspection, dynamic
Enhancing scanning is conventional to divide three phases i.e. arterial phase, venous phase, balance period.
Pancreatic Neuroendocrine Tumors:Pancreatic Neuroendocrine Tumors (pancreatic
Neuroendocrineneoplasm, pNENs), original is known as islet-cell tumour, accounts for about the 3% of primary pancreatic neoplasm.According to sharp
The clinical manifestation of the secretor state and patient of element, Pancreatic Neuroendocrine Tumors are divided into functional pNENs and non-functional
pNENs。
The classification of tumour:Mainly according to the differentiation degree of tumour cell.Atypia and nuclear fission number determine, are generally divided into
Three-level, G1:Well differentiated, nuclear fission is rare, belongs to low potential malignancy;G2:Break up medium, nuclear fission is clear to, and it is pernicious to belong to moderate;G3:
Break up poor, nuclear fission is more, belongs to high malignancy.
Deep learning:Deep learning is substantially exactly multilayer neural network, and one layer of neural network contains one linearly
Transformation adds a nonlinear operation, and multilayer neural network is exactly the compound of multiple nonlinear functions in fact.Deep learning can
More abstract high-rise expression attribute classification or feature are formed by combining low-level feature, to find the distributed nature table of data
Show.
Convolutional neural networks:Convolutional neural networks (Convolutional Neural Network, CNN) are a kind of feedforwards
Neural network, its artificial neuron can respond the surrounding cells in a part of coverage area, have for large-scale image procossing
Outstanding performance.It includes convolutional layer (Convolutional layer), pond layer (Pooling layer) and full articulamentum
(Fully connected layer)。
For the treatment of Pancreatic Neuroendocrine Tumors, according to the grade malignancy of tumour height, different treatments is had
Scheme can learn the grade malignancy height of tumour by the classification of tumour.For the pathological grading of pancreas neuroendocrine tumors
Or grade malignancy, it is merely able to make a definite diagnosis by pathologic finding at present, pathologic finding mainly has two kinds of operation excision and living body puncture
Approach:Operation excision is carried out for tumour that is not spreading or only being spread in a small range, then again it is organized to carry out
Pathologic finding determines the grade malignancy of tumour;For having there is the tumour spread to a certain degree, pathology is done by living body puncture
Classification, determines its grade malignancy.Pancreas neuroendocrine tumors onset initial symptoms are hidden, and just there is symptom in late period, so when first visit
There is 65% or so displaced, therefore the diagnosis of the grade malignancy of most tumour is all to be made a definite diagnosis by living body puncture and really
Determine therapeutic scheme.The treatment of tumour is a prolonged process, and over the course for the treatment of, the grade malignancy of tumour may be because of
Certain factors and change, can be lossless, detect malignancy in real time over the course for the treatment of by diagnostic imaging
Variation, guidance when need to be punctured again, avoid it is unnecessary puncture patient is caused to damage.
Based on CT images to the diagnostic techniques of the grade malignancy of Pancreatic Neuroendocrine Tumors, what is relied at present is image department
Doctor's visually observes assessment, artificial diagosis.The diagnostic imaging mode of existing this artificial diagosis, dependent on image doctor's
Personal experience, the unified standard of neither one, often different doctors even diagnosis knot of the same doctor to the image of same patient
Fruit can have differences, meanwhile, the diagosis work of image department doctor belongs to prolonged repeated labor, and Yi Yinqi doctor's is tired
Labor leads to repetitive stain injury.
In conclusion in the industry there is an urgent need for it is a kind of it is objective, accurate and efficient Pancreatic Neuroendocrine Tumors grade malignancy is true
Determine scheme.
Invention content
In order to solve the above technical problems, it is an object of the invention to:A kind of objective, accurate and efficient pancreas god is provided
Computer diagnosis method and system through endocrine tumors grade malignancy.
The first technical solution for being taken of the present invention is:
The computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy, includes the following steps:
Initial data is obtained, the initial data includes the patient's image data and patient's disease of Pancreatic Neuroendocrine Tumors
Manage ranked data;
Initial data is divided as unit of patient, obtains training set, test set and verification collection;
It is trained and tests using convolutional neural networks according to training set, test set and verification collection, obtain pancreas nerve
Endocrine tumors grade malignancy diagnostic model;
Patient data to be diagnosed is inputted into Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, exports corresponding pancreas
Gland neuroendocrine tumor grade malignancy.
Further, described that initial data is divided as unit of patient, it obtains training set, test set and verification and collects this
One step, specifically includes:
Initial data is pre-processed, data after being pre-processed;
Data after pretreatment are divided as unit of patient, obtain training set, test set and verification collection.
Further, described that initial data is pre-processed, after being pre-processed the step for data, specifically include:
Tumor focus extraction is carried out to initial data, obtains the corresponding three-dimensional matrice in the region containing tumor section;
Obtained three-dimensional matrice is normalized, the three-dimensional matrice after being normalized;
The three-dimensional matrice after normalization is sampled along z-axis direction, obtains sample, wherein z-axis characterization contains tumour
Image layer the number of plies, the three-dimensional matrice after normalization is divided into several submatrixs by sampling, and while sampling adjacent two sons
There are 50% intersections for matrix.
Further, described to be trained and test using convolutional neural networks according to training set, test set and verification collection, it obtains
It the step for Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, specifically includes:
Data amplification, training set after being expanded are carried out to training set;
It is trained using convolutional neural networks according to training set after amplification, the pancreas neuroendocrine after being trained is swollen
Tumor grade malignancy diagnostic model;
Model discrimination is carried out using the Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model after verification set pair training, is obtained
To final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test is carried out to final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model using test set,
Export corresponding model performance test result.
Further, the convolutional neural networks use the three-dimensional residual error network based on ResNet models.
Further, the convolutional neural networks include 1 Three dimensional convolution layer, 1 three-dimensional maximum value pond layer, 12 first
Block units and 4 the 2nd Block units, wherein the first Block units include the first convolutional layer, the second convolutional layer, third
Convolutional layer and active coating, the input of input the first Block units of connection of the first convolutional layer, the output of the first convolutional layer is by swashing
Layer living is connect with the input of the second convolutional layer, and the output of the second convolutional layer is connect by active coating with the input of third convolutional layer,
The output of third convolutional layer and output of the connection active coating as the first Block units after the input summation of the first Block units;
2nd Block units include third convolutional layer, Volume Four lamination, the 5th convolutional layer, the 6th convolutional layer and active coating, third convolution
The input of layer and the input of the 6th convolutional layer are all connected with the input of the 2nd Block units, and the output of third convolutional layer passes through activation
Layer is connect with the input of Volume Four lamination, and the output of Volume Four lamination is connect by active coating with the input of the 5th convolutional layer, the
The output of five convolutional layers and output of the connection active coating as the 2nd Block units after the output summation of the 6th convolutional layer.
Further, the method that model performance test uses is eight folding cross-validation methods, the evaluation index used for by
Examination person's performance curve.
The second technical solution for being taken of the present invention is:
The computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy, including:
Acquisition module, for obtaining initial data, the initial data includes patient's shadow of Pancreatic Neuroendocrine Tumors
As data and patient's pathological grading data;
Division module obtains training set, test set and verification for being divided to initial data as unit of patient
Collection;
Training and test module, for being trained using convolutional neural networks according to training set, test set and verification collection
And test, obtain Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Computer diagnosis module is examined for patient data to be diagnosed to be inputted Pancreatic Neuroendocrine Tumors grade malignancy
Disconnected model, exports corresponding Pancreatic Neuroendocrine Tumors grade malignancy.
Further, the training includes with test module:
Data amplification unit, for carrying out data amplification, training set after being expanded to training set;
Training unit, for being trained using convolutional neural networks according to training set after amplification, the pancreas after being trained
Gland neuroendocrine tumor grade malignancy diagnostic model;
Model discrimination unit, for diagnosing mould using the Pancreatic Neuroendocrine Tumors grade malignancy after verification set pair training
Type carries out model discrimination, obtains final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test cell, for being diagnosed to final Pancreatic Neuroendocrine Tumors grade malignancy using test set
Model carries out model performance test, exports corresponding model performance test result.
The third technical solution taken of the present invention is:
The computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy, including:
Memory, for storing program;
Processor, for loading described program to execute the computer diagnosis method as described in the first technical solution.
The beneficial effects of the invention are as follows:The computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention and
System is carried out data division as unit of patient, and uses convolutional neural networks to carry out Pancreatic Neuroendocrine Tumors evil
Property degree diagnostic model training and test, finally can by Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model automatically according to
Patient data to be diagnosed exports corresponding grade malignancy, is realized by deep learning pernicious to Pancreatic Neuroendocrine Tumors
The computer diagnosis of degree can objectively assess the image of patient, keep the result of diagnostic imaging more objective and accurate, and one
Determine to reduce the probability of the young doctor's mistaken diagnosis lacked experience in degree, while the diagosis that can reduce image department doctor is born, and is carried
The working efficiency of Gao doctors.
Description of the drawings
Fig. 1 is the step flow chart of the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention;
Fig. 2 is that the present invention is based on the methods of deep learning to establish Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model
Particular flow sheet;
Fig. 3 is a certain layer original CT images;
Fig. 4 is images of the Fig. 3 after doctor marks;
Fig. 5 is the front and back image array size variation schematic diagram of patient data's mark;
Fig. 6 is the process schematic that the present invention samples;
Fig. 7 is a kind of concrete structure schematic diagram of residual error network of the present invention;
Fig. 8 is the structural schematic diagram of the first Block units;
Fig. 9 is the structural schematic diagram of the 2nd Block units;
Figure 10 is the ROC curve figure of the eight folding cross validation of diagnostic model of the present invention;
Figure 11 is the diagnostic model of the present invention according to the ROC curve figure of data center's training test;
Figure 12 is the diagnostic model of the present invention according to the ROC curve figure of timing node partition testing.
Specific implementation mode
Referring to Fig.1, the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention, including following step
Suddenly:
Initial data is obtained, the initial data includes the patient's image data and patient's disease of Pancreatic Neuroendocrine Tumors
Manage ranked data;
Initial data is divided as unit of patient, obtains training set, test set and verification collection;
It is trained and tests using convolutional neural networks according to training set, test set and verification collection, obtain pancreas nerve
Endocrine tumors grade malignancy diagnostic model;
Patient data to be diagnosed is inputted into Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, exports corresponding pancreas
Gland neuroendocrine tumor grade malignancy.
It is further used as preferred embodiment, it is described that initial data is divided as unit of patient, it is trained
It the step for collection, test set and verification collection, specifically includes:
Initial data is pre-processed, data after being pre-processed;
Data after pretreatment are divided as unit of patient, obtain training set, test set and verification collection.
Be further used as preferred embodiment, it is described that initial data is pre-processed, after being pre-processed data this
One step, specifically includes:
Tumor focus extraction is carried out to initial data, obtains the corresponding three-dimensional matrice in the region containing tumor section;
Obtained three-dimensional matrice is normalized, the three-dimensional matrice after being normalized;
The three-dimensional matrice after normalization is sampled along z-axis direction, obtains sample, wherein z-axis characterization contains tumour
Image layer the number of plies, the three-dimensional matrice after normalization is divided into several submatrixs by sampling, and while sampling adjacent two sons
There are 50% intersections for matrix.
Wherein, tumor focus extraction can be screened by modes such as doctor's marks from initial data.
And it normalizes and z-score normalization functions can be used to realize.
It is further used as preferred embodiment, it is described that convolutional Neural net is used according to training set, test set and verification collection
Network is trained and tests, the step for obtaining Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, specifically includes:
Data amplification, training set after being expanded are carried out to training set;
It is trained using convolutional neural networks according to training set after amplification, the pancreas neuroendocrine after being trained is swollen
Tumor grade malignancy diagnostic model;
Model discrimination is carried out using the Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model after verification set pair training, is obtained
To final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test is carried out to final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model using test set,
Export corresponding model performance test result.
Wherein, the method for data amplification adds including rotation, setting contrast and image makes an uproar.
It is further used as preferred embodiment, the convolutional neural networks use the three-dimensional residual error based on ResNet models
Network.
Be further used as preferred embodiment, the convolutional neural networks include 1 Three dimensional convolution layer, 1 three-dimensional most
Big value pond layer, 12 the first Block units and 4 the 2nd Block units, wherein the first Block units include the first convolution
Layer, the second convolutional layer, third convolutional layer and active coating, the input of input the first Block units of connection of the first convolutional layer, first
The output of convolutional layer is connect by active coating with the input of the second convolutional layer, and the output of the second convolutional layer passes through active coating and third
The input of convolutional layer connects, and the output of third convolutional layer and connection active coating after the input summation of the first Block units are as the
The output of one Block units;2nd Block units include third convolutional layer, Volume Four lamination, the 5th convolutional layer, the 6th convolution
Layer and active coating, the input of third convolutional layer and the input of the 6th convolutional layer are all connected with the input of the 2nd Block units, third volume
The output of lamination is connect by active coating with the input of Volume Four lamination, and the output of Volume Four lamination passes through active coating and volume five
The input of lamination connects, and the output of the 5th convolutional layer is with connection active coating after the output summation of the 6th convolutional layer as second
The output of Block units.
It is further used as preferred embodiment, the method that the model performance test uses is eight folding cross-validation methods,
The evaluation index used is Receiver operating curve.
It is corresponding with the method for Fig. 1, the computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention, packet
It includes:
Acquisition module, for obtaining initial data, the initial data includes patient's shadow of Pancreatic Neuroendocrine Tumors
As data and patient's pathological grading data;
Division module obtains training set, test set and verification for being divided to initial data as unit of patient
Collection;
Training and test module, for being trained using convolutional neural networks according to training set, test set and verification collection
And test, obtain Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Computer diagnosis module is examined for patient data to be diagnosed to be inputted Pancreatic Neuroendocrine Tumors grade malignancy
Disconnected model, exports corresponding Pancreatic Neuroendocrine Tumors grade malignancy.
It is further used as preferred embodiment, the training includes with test module:
Data amplification unit, for carrying out data amplification, training set after being expanded to training set;
Training unit, for being trained using convolutional neural networks according to training set after amplification, the pancreas after being trained
Gland neuroendocrine tumor grade malignancy diagnostic model;
Model discrimination unit, for diagnosing mould using the Pancreatic Neuroendocrine Tumors grade malignancy after verification set pair training
Type carries out model discrimination, obtains final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test cell, for being diagnosed to final Pancreatic Neuroendocrine Tumors grade malignancy using test set
Model carries out model performance test, exports corresponding model performance test result.
It is corresponding with the method for Fig. 1, the computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention, packet
It includes:
Memory, for storing program;
Processor, for loading described program to execute computer diagnosis method as described in the present invention.
The present invention is further explained and is illustrated with specific embodiment with reference to the accompanying drawings of the specification.
It is deposited by way of grade malignancy of the artificial diagosis to determine Pancreatic Neuroendocrine Tumors for the prior art
Defect, the present embodiment proposes a kind of computer diagnosis scheme of gland neuroendocrine tumor grade malignancy, and the program is first
Method based on deep learning establishes Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, then can according to the model come
The automatic diagnosis for carrying out Pancreatic Neuroendocrine Tumors grade malignancy.As shown in Fig. 2, method of the program based on deep learning is built
Vertical Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model specifically includes the following contents:
One, data collection and data prediction
The present embodiment have collected Pancreatic Neuroendocrine Tumors patient's image and its pathological grading data as original number
According to using the abdominal CT enhanced images of arterial phase, and its pathological grading result must be that clinically final is made a definite diagnosis
As a result.As shown in figure 3, the original patient image data is three-dimensional abdominal CT images, each layer of image size is 512 ×
512, i.e. x-axis and y-axis direction size is 512, and the image number of plies (i.e. the size in z-axis direction) having per an example patient from
176 to 588 etc., wherein including that the image layer number of plies of tumour is differed 10 to 200 in each patient image.Such as Fig. 4 and Fig. 5
It is shown, can tumour be first marked manually with rectangle frame by doctor, will contain tumour portion in image as unit of patient in pretreatment
The rectangular area divided splits and deposits in corresponding three-dimensional matrice, and the length and width, that is, x-axis and y-axis size of the three-dimensional matrice are equal
It is 256, high (i.e. z-axis) is the number of plies of the image layer containing tumour in the 3-D view for correspond to patient;
Then operation is normalized to the three-dimensional matrice extracted, normalization mode is that 0 mean value standardizes (Z-
Score standardization), formula is as follows:
Wherein, μ and σ is respectively the mean value and variance of raw data set.
The matrix after normalization is sampled along z-axis direction again, to which a complete tumour is divided into multiple 256
× 256 × 16 submatrix.As shown in fig. 6, two submatrixs adjacent when sampling have 50% intersection, make a disease
People has the image data of plural number, to expand sample size;
Then according to patient by sample be divided into training set, verification collection and test set, by rotation, setting contrast with
And image adds and makes an uproar, and training set data is carried out data amplification.The patient's sample that pathological grading is G1 and G2 can be wherein expanded to
About 10,000 altogether, the patient's sample of G3 is expanded to about 10,000, i.e., about 10,000, the sample of low grade malignancy, the sample of high grade malignancy
This about 10,000.
In subsequent step, network model is trained by training set, with verification collection come preference pattern, with test set come pair
Selected diagnostic model test, assesses the performance of diagnostic model.
Two, network structure designs
Residual error network is easier to optimize, and can improve accuracy rate by increasing corresponding depth, solves increase
The side effect (i.e. degenerate problem) that depth zone comes, in this way can be by merely increasing network depth, to improve network performance.Therefore
The present embodiment has selected residual error network to build the network structure of convolutional neural networks.The present embodiment is with reference to residual error network ResNet
The network structure of middle Res50 devises the Three dimensional convolution neural network suitable for the pretreated patient data of step 1, network
Involved parameter in design process can be transferred through the test result of training set and verification collection data to adjust and optimize.
The Three dimensional convolution neural network of the present embodiment include 1 Three dimensional convolution layer, 1 three-dimensional maximum value pond layer, 12
First Block units and 4 the 2nd Block units, wherein the first Block units include the first convolutional layer, the second convolutional layer,
Third convolutional layer and active coating, the input of input the first Block units of connection of the first convolutional layer, the output of the first convolutional layer are logical
It crosses active coating to connect with the input of the second convolutional layer, the output of the second convolutional layer is connected by active coating and the input of third convolutional layer
It connects, the output of third convolutional layer is with connection active coating after the input summation of the first Block units as the defeated of the first Block units
Go out;2nd Block units include third convolutional layer, Volume Four lamination, the 5th convolutional layer, the 6th convolutional layer and active coating, third
The input of convolutional layer and the input of the 6th convolutional layer are all connected with the input of the 2nd Block units, and the output of third convolutional layer passes through
Active coating is connect with the input of Volume Four lamination, and the output of Volume Four lamination is connected by active coating and the input of the 5th convolutional layer
It connects, the output of the 5th convolutional layer and output of the connection active coating as the 2nd Block units after the output summation of the 6th convolutional layer.
The sequencing of 18 layers of structure is:It is mono- to input → 1 Three dimensional convolution layer → 1 three-dimensional maximum value pond the 2nd Block of layer → 1
Member → 2 sequentially connected first Block units in sequentially connected the 2nd unit → 3 Block of first unit → 1st Block
Sequentially connected the 2nd unit → 2 Block of first unit → 1st Block in → 1 the 2nd unit → 5 Block are sequentially connected
The first Block units.
The main function of Three dimensional convolution layer and the first to the 6th convolutional layer is carried out to essential characteristic by convolution operation
Permutation and combination, feature with semantic information more abstract to obtain.Active coating can increase the non-thread of convolutional neural networks
Property, be conducive to convolutional neural networks convergence.Active coating can be selected rectification linear unit, sigmoid functions etc. and be used as activation primitive.
Preferably, the convergence rate that the linear unit function of rectification accelerates convolutional neural networks as activation primitive can be selected in active coating.
Three-dimensional maximum value pond layer, the length for reducing input feature vector figure and width, its significance lies in that Connecting quantity and calculation amount are reduced, with
Meet shift invariant and obtains information more of overall importance.Because with the constant filtering of size on the figure after the diminution of pond layer
Device, it is meant that the opposite local receptor field of each neuron can become larger, and next layer of each neuron is enable to extract more entirely
The feature of office's property.First Block units and the 2nd Block units can be (quick by feedforward neural network+shortcut connections
Connection, input jump directly to a certain layer or output) it realizes, as shown in Figure 8 and Figure 9.Shortcut connections are equivalent to simple execution
Same mapping, not will produce additional parameter, will not increase computation complexity, moreover, whole network can still pass through
Backpropagation end to end is trained.
Several width can be obtained after above-mentioned 18 layers of structure in the image (part of pretreated patient data) of input
Characteristic pattern is as output.This several width characteristic pattern using after global mean value pondization and sigmoid function Nonlinear Processings i.e.
The anticipation function of diagnostic model can be generated.
Three, the foundation, screening and assessment of diagnostic model
In the present embodiment, patient's image that pathological grading is G1 and G2 is set to low grade malignancy as the first kind, disease
Patient's image that reason is classified as G3 is set to high grade malignancy as the second class, is obtained with the training set data training network after amplification
Two sorter network models, when choosing optimal diagnosis model and its performance evaluating, respectively with without the verification collection of amplification and survey
Examination collects data to carry out, and need to be used as the input of model as unit of patient, is exported to the patient's according to the mechanism of ballot
Diagnostic result has been broken down into N number of data, and have examining for n1 data in a model if after a patient samples by pressing z-axis
Disconnected result is the first kind, and the diagnostic result of n2 data is the second class, n1+n2=N;As n2 >=n1, the prediction to the patient
As a result be the second class, on the contrary it is then be the first kind, be achieved in the diagnosis to patient's grade malignancy height.
The present embodiment uses Receiver operating curve (receiver operating characteristic
Curve, abbreviation ROC curve) assessment disaggregated model performance, the test result by calculating all subjects draws corresponding classification
The ROC curve of model.In K rolls over cross validation, all subjects refer to the set of K test set, with the face under ROC curve
Accuracy rate, susceptibility and spy under product (Area under curve, AUC), and the optimal classification threshold value that is determined by ROC curve
Different degree assesses the performance of disaggregated model, wherein accuracy rate Accuracy, susceptibility Sensitivity, specificity
The computational methods of Specificity are as follows:
Accuracy=(TP+TN)/(TP+TN+FP+FN)
Sensitivity=TP/ (TP+FN)
Specificity=TN/ (TN+FP)
Wherein, TP:True Positive are judged as positive sample, are in fact also the total sample number of positive sample.
TN:True Negative are judged as negative sample, are in fact also the total sample number of negative sample.
FP:False Positive are judged as positive sample, but are in fact the total sample numbers of negative sample.
FN:False Negative are judged as negative sample, but are in fact the total sample numbers of positive sample.
Scheme in order to better illustrate the present invention is illustrated with reference to specific application example:
It is collected from Liang Ge hospitals and comes totally 103 patient datas, wherein G1:40, G2:44, G3:19, pass through this
The technical solution of invention is pre-processed and is modeled to data, the z-axis in technical solution pre-treatment step through the invention
The quantity of sample is expanded to 814, wherein G1 by the direction method of sampling:172, G2:394, G3:248.
With reference to the network structure of Res50 in residual error network ResNet, the residual error network of 3 dimensions is specially devised to carry out tumour
The concrete structure of the diagnosis of grade malignancy, the residual error network is as shown in Figure 7.
It is described as follows accordingly in Fig. 7:
ImageSize:The size and number of the characteristic pattern of output.For example, 128 × 128 × 16@64 indicate the big of characteristic pattern
Small is 128 × 128 × 16, shares 64 groups of characteristic patterns.
Conv3D:3 dimension convolution, parameter such as (3 × 3 × 3,64), indicate the size of convolution kernel be 3 × 3 × 3,64 be roll up
The quantity of the characteristic pattern of image after product core group number or convolution.
Strides:Indicate the step-length that convolution kernel moves every time in three dimensions.Such as:Strides=(3,2,1) table
Show that in each moving step length of convolution kernel in the direction of the x axis be 3, each moving step length in y-axis direction is 2, in z-axis direction
Each moving step length is 1.
MaxPooling3D:Three-dimensional maximum value pond, 3 × 3 × 3 indicate pond when convolution kernel size, Strides=(2,
2,1) step-length of convolution kernel movement is indicated.
GlobalAveragePooling:Global mean value pond, the mean value for characteristic pattern to be carried out to whole characteristic pattern
Chi Hua.One characteristic pattern forms a characteristic point behind global mean value pond, and 2048 width characteristic patterns can form 2048 spies
Sign point.
prediction:The anticipation function of diagnostic model.
And the concrete structure of the first Block unit Bs lock1 as shown in figure 8, the 2nd Block unit Bs lock2 specific knot
Structure is as shown in Figure 9.
Then, following three kinds different training test methods are respectively adopted to verify network structure used in this example for pancreas
The diagnosis effect of gland neuroendocrine tumor grade malignancy:
The training test of (1) eight folding cross validation
In eight folding cross validations, total data is divided into 8 foldings, includes the figure of complete 13 patients in each folding
As data, preceding four compromise, each folding has G1:5, G2:6, G3:2;Four compromises, each folding have G1 afterwards:5, G2:5, G3:3;Each time
It takes a wherein folding as test set, then collects as verification from remaining seven broken number according to 6 patients are randomly selected in totally 90 patients,
Remaining 84 patients expand training set data as training set, are used in combination the training network of the data after amplification to obtain diagnostic model, lead to
Cross verification collection and examine, choose diagnostic model, finally with select come model test set is diagnosed.
By 8 broken numbers according to alternately as test set, the diagnostic result to all patients is obtained, to obtain examining for network used
Disconnected effect.When carrying out performance evaluating, the ROC curve based on Figure 10, specific performance evaluation results, which can be obtained, includes:AUC:
0.79, under optimal classification threshold value, accuracy rate:80.58%, susceptibility:78.95%, specificity:80.95%.
(2) it trains and tests according to data center
Trained according to data center test when, using the data collected of certain famous hospitals of the country as training set and from
In randomly selected 6 patients and collected as verification, the data that another hospital collects are as test set;That is training set totally 78
Example patient, verification collection totally 6 patients, test set totally 19 patients, wherein G1:6, G2:7, G3:6;The test of this method
Obtained AUC:0.82, under optimal classification threshold value, accuracy rate:84.21%, susceptibility:83.33%, specificity:84.21%,
ROC curve is as shown in figure 11.
(3) according to timing node partition testing
When according to timing node partition testing, using the data before in June, 2016 as training set, after in June, 2016
Data be that verification collects, 6 patients are equally randomly selected from training set and are collected as verification;That is training set totally 74, verification collection 6
Example, test set totally 23, wherein G1:6, G2:11, G3:6;The AUC that the test of this method obtains:0.82, most preferably dividing
Under class threshold value, accuracy rate:72.73%, susceptibility:66.67%, specificity:75%, ROC curve is as shown in figure 12.
By Figure 10, Figure 11 and Figure 12 and corresponding test result it is found that the network structure swells for pancreas neuroendocrine
The diagnosis capability of tumor grade malignancy is preferable.
In conclusion the computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy of the present invention, are based on
Convolutional neural networks carry out data division as unit of patient, are realized in pancreas nerve points by the method for deep learning
The computer diagnosis for secreting malignancy, error caused by reducing the subjective factor that Artificial Diagnosis is brought, allows doctor can
The image for more objectively assessing patient, keeps the result of diagnostic imaging more objective, accurate, reduces young shortage warp to a certain extent
The probability for the doctor's mistaken diagnosis tested, while the diagosis burden of image department doctor can be reduced, improve the working efficiency of doctors.
It is to be illustrated to the preferable implementation of the present invention, but the present invention is not limited to the embodiment above, it is ripe
Various equivalent variations or replacement can also be made under the premise of without prejudice to spirit of that invention by knowing those skilled in the art, this
Equivalent deformation or replacement are all contained in the application claim limited range a bit.
Claims (10)
1. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy, it is characterised in that:Include the following steps:
Initial data is obtained, the initial data includes the patient's image data and patient's pathology point of Pancreatic Neuroendocrine Tumors
Level data;
Initial data is divided as unit of patient, obtains training set, test set and verification collection;
It is trained and tests using convolutional neural networks according to training set, test set and verification collection, obtain dividing in pancreas nerve
Secrete malignancy diagnostic model;
Patient data to be diagnosed is inputted into Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model, exports corresponding pancreas god
Through endocrine tumors grade malignancy.
2. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 1, feature exist
In:It is described that initial data is divided as unit of patient, the step for training set, test set and verification collect is obtained, specifically
Including:
Initial data is pre-processed, data after being pre-processed;
Data after pretreatment are divided as unit of patient, obtain training set, test set and verification collection.
3. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 2, feature exist
In:It is described that initial data is pre-processed, after being pre-processed the step for data, specifically include:
Tumor focus extraction is carried out to initial data, obtains the corresponding three-dimensional matrice in the region containing tumor section;
Obtained three-dimensional matrice is normalized, the three-dimensional matrice after being normalized;
The three-dimensional matrice after normalization is sampled along z-axis direction, obtains sample, wherein z-axis characterizes the figure containing tumour
As the number of plies of layer, the three-dimensional matrice after normalization is divided into several submatrixs, and adjacent two submatrixs when sampling by sampling
There are 50% intersections.
4. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 1, feature exist
In:It is described to be trained and test using convolutional neural networks according to training set, test set and verification collection, it obtains in pancreas nerve
It the step for secreting tumor grade malignancy diagnostic model, specifically includes:
Data amplification, training set after being expanded are carried out to training set;
It is trained using convolutional neural networks according to training set after amplification, the Pancreatic Neuroendocrine Tumors after being trained are disliked
Property degree diagnostic model;
Model discrimination is carried out using the Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model after verification set pair training, is obtained most
Whole Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test, output are carried out to final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model using test set
Corresponding model performance test result.
5. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 1, feature exist
In:The convolutional neural networks use the three-dimensional residual error network based on ResNet models.
6. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 5, feature exist
In:The convolutional neural networks include 1 Three dimensional convolution layer, 1 three-dimensional maximum value pond layer, 12 the first Block units and 4
A 2nd Block units, wherein the first Block units include the first convolutional layer, the second convolutional layer, third convolutional layer and activation
Layer, the input of input the first Block units of connection of the first convolutional layer, the output of the first convolutional layer pass through active coating and volume Two
The input of lamination connects, and the output of the second convolutional layer is connect by active coating with the input of third convolutional layer, third convolutional layer
Output and output of the connection active coating as the first Block units after the input summation of the first Block units;2nd Block is mono-
Member include third convolutional layer, Volume Four lamination, the 5th convolutional layer, the 6th convolutional layer and active coating, the input of third convolutional layer and
The input of 6th convolutional layer is all connected with the input of the 2nd Block units, and the output of third convolutional layer passes through active coating and Volume Four
The input of lamination connects, and the output of Volume Four lamination is connect by active coating with the input of the 5th convolutional layer, the 5th convolutional layer
Output and output of the connection active coating as the 2nd Block units after the output summation of the 6th convolutional layer.
7. the computer diagnosis method of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 4, feature exist
In:The method that the model performance test uses is eight folding cross-validation methods, and the evaluation index used is Receiver Operating Characteristics
Curve.
8. the computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy, it is characterised in that:Including:
Acquisition module, for obtaining initial data, the initial data includes patient's image number of Pancreatic Neuroendocrine Tumors
According to patient's pathological grading data;
Division module obtains training set, test set and verification collection for being divided to initial data as unit of patient;
Training and test module, for being trained and surveying using convolutional neural networks according to training set, test set and verification collection
Examination, obtains Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Computer diagnosis module, for patient data's input Pancreatic Neuroendocrine Tumors grade malignancy diagnosis mould that will be to be diagnosed
Type exports corresponding Pancreatic Neuroendocrine Tumors grade malignancy.
9. the computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy according to claim 8, feature exist
In:The training includes with test module:
Data amplification unit, for carrying out data amplification, training set after being expanded to training set;
Training unit, for being trained using convolutional neural networks according to training set after amplification, the pancreas god after being trained
Through endocrine tumors grade malignancy diagnostic model;
Model discrimination unit, for using verification set pair training after Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model into
Row model discrimination obtains final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model;
Model performance test cell, for using test set to final Pancreatic Neuroendocrine Tumors grade malignancy diagnostic model
Model performance test is carried out, corresponding model performance test result is exported.
10. the computerized diagnostic system of Pancreatic Neuroendocrine Tumors grade malignancy, it is characterised in that:Including:
Memory, for storing program;
Processor, for loading described program to execute such as claim 1-7 any one of them computer diagnosis methods.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810306839.6A CN108766555A (en) | 2018-04-08 | 2018-04-08 | The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810306839.6A CN108766555A (en) | 2018-04-08 | 2018-04-08 | The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108766555A true CN108766555A (en) | 2018-11-06 |
Family
ID=63981166
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810306839.6A Pending CN108766555A (en) | 2018-04-08 | 2018-04-08 | The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108766555A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110047082A (en) * | 2019-03-27 | 2019-07-23 | 深圳大学 | Pancreatic Neuroendocrine Tumors automatic division method and system based on deep learning |
CN110910371A (en) * | 2019-11-22 | 2020-03-24 | 北京理工大学 | Liver tumor automatic classification method and device based on physiological indexes and image fusion |
CN111292304A (en) * | 2020-01-22 | 2020-06-16 | 青岛大学附属医院 | Pancreatic tumor pathology automatic diagnosis system based on deep neural network |
CN111429968A (en) * | 2020-03-11 | 2020-07-17 | 至本医疗科技(上海)有限公司 | Method, electronic device, and computer storage medium for predicting tumor type |
CN111584071A (en) * | 2020-05-11 | 2020-08-25 | 何方舟 | Artificial intelligent auxiliary diagnosis model for primary malignant bone tumor around knee joint |
CN112700859A (en) * | 2020-12-15 | 2021-04-23 | 贵州小宝健康科技有限公司 | Medical diagnosis assisting method and system based on medical images |
CN113299385A (en) * | 2021-04-30 | 2021-08-24 | 北京深睿博联科技有限责任公司 | Pancreatic cystic lesion clinical decision method and system based on deep learning |
CN114677378A (en) * | 2022-05-31 | 2022-06-28 | 四川省医学科学院·四川省人民医院 | Computer-aided diagnosis and treatment system based on ovarian tumor benign and malignant prediction model |
CN116705252A (en) * | 2023-06-16 | 2023-09-05 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105117611A (en) * | 2015-09-23 | 2015-12-02 | 北京科技大学 | Determining method and system for traditional Chinese medicine tongue diagnosis model based on convolution neural networks |
US20170140260A1 (en) * | 2015-11-17 | 2017-05-18 | RCRDCLUB Corporation | Content filtering with convolutional neural networks |
CN106780498A (en) * | 2016-11-30 | 2017-05-31 | 南京信息工程大学 | Based on point depth convolutional network epithelium and matrix organization's automatic division method pixel-by-pixel |
CN107169974A (en) * | 2017-05-26 | 2017-09-15 | 中国科学技术大学 | It is a kind of based on the image partition method for supervising full convolutional neural networks more |
US9836853B1 (en) * | 2016-09-06 | 2017-12-05 | Gopro, Inc. | Three-dimensional convolutional neural networks for video highlight detection |
CN107451609A (en) * | 2017-07-24 | 2017-12-08 | 上海交通大学 | Lung neoplasm image identification system based on depth convolutional neural networks |
CN107590797A (en) * | 2017-07-26 | 2018-01-16 | 浙江工业大学 | A kind of CT images pulmonary nodule detection method based on three-dimensional residual error neutral net |
-
2018
- 2018-04-08 CN CN201810306839.6A patent/CN108766555A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105117611A (en) * | 2015-09-23 | 2015-12-02 | 北京科技大学 | Determining method and system for traditional Chinese medicine tongue diagnosis model based on convolution neural networks |
US20170140260A1 (en) * | 2015-11-17 | 2017-05-18 | RCRDCLUB Corporation | Content filtering with convolutional neural networks |
US9836853B1 (en) * | 2016-09-06 | 2017-12-05 | Gopro, Inc. | Three-dimensional convolutional neural networks for video highlight detection |
CN106780498A (en) * | 2016-11-30 | 2017-05-31 | 南京信息工程大学 | Based on point depth convolutional network epithelium and matrix organization's automatic division method pixel-by-pixel |
CN107169974A (en) * | 2017-05-26 | 2017-09-15 | 中国科学技术大学 | It is a kind of based on the image partition method for supervising full convolutional neural networks more |
CN107451609A (en) * | 2017-07-24 | 2017-12-08 | 上海交通大学 | Lung neoplasm image identification system based on depth convolutional neural networks |
CN107590797A (en) * | 2017-07-26 | 2018-01-16 | 浙江工业大学 | A kind of CT images pulmonary nodule detection method based on three-dimensional residual error neutral net |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110047082A (en) * | 2019-03-27 | 2019-07-23 | 深圳大学 | Pancreatic Neuroendocrine Tumors automatic division method and system based on deep learning |
CN110910371A (en) * | 2019-11-22 | 2020-03-24 | 北京理工大学 | Liver tumor automatic classification method and device based on physiological indexes and image fusion |
CN111292304A (en) * | 2020-01-22 | 2020-06-16 | 青岛大学附属医院 | Pancreatic tumor pathology automatic diagnosis system based on deep neural network |
CN111429968A (en) * | 2020-03-11 | 2020-07-17 | 至本医疗科技(上海)有限公司 | Method, electronic device, and computer storage medium for predicting tumor type |
CN111584071A (en) * | 2020-05-11 | 2020-08-25 | 何方舟 | Artificial intelligent auxiliary diagnosis model for primary malignant bone tumor around knee joint |
CN112700859A (en) * | 2020-12-15 | 2021-04-23 | 贵州小宝健康科技有限公司 | Medical diagnosis assisting method and system based on medical images |
CN113299385A (en) * | 2021-04-30 | 2021-08-24 | 北京深睿博联科技有限责任公司 | Pancreatic cystic lesion clinical decision method and system based on deep learning |
CN113299385B (en) * | 2021-04-30 | 2024-03-08 | 北京深睿博联科技有限责任公司 | Deep learning-based pancreatic cystic disease clinical decision method and system |
CN114677378A (en) * | 2022-05-31 | 2022-06-28 | 四川省医学科学院·四川省人民医院 | Computer-aided diagnosis and treatment system based on ovarian tumor benign and malignant prediction model |
CN114677378B (en) * | 2022-05-31 | 2022-09-02 | 四川省医学科学院·四川省人民医院 | Computer-aided diagnosis and treatment system based on ovarian tumor benign and malignant prediction model |
CN116705252A (en) * | 2023-06-16 | 2023-09-05 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
CN116705252B (en) * | 2023-06-16 | 2024-05-31 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108766555A (en) | The computer diagnosis method and system of Pancreatic Neuroendocrine Tumors grade malignancy | |
van Hamersvelt et al. | Deep learning analysis of left ventricular myocardium in CT angiographic intermediate-degree coronary stenosis improves the diagnostic accuracy for identification of functionally significant stenosis | |
CN111243042A (en) | Ultrasonic thyroid nodule benign and malignant characteristic visualization method based on deep learning | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
Alksas et al. | A novel computer-aided diagnostic system for accurate detection and grading of liver tumors | |
JP2004174254A (en) | Method and system for measuring disease related tissue change | |
CN105934199A (en) | Apparatus and method for processing a medical image of a body lumen | |
CN108209955A (en) | The characterization of patch | |
CN110021022A (en) | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning | |
Pan et al. | Detecting the early infarct core on non-contrast CT images with a deep learning residual network | |
Kawasaki et al. | Evaluation of significant coronary artery disease based on CT fractional flow reserve and plaque characteristics using random forest analysis in machine learning | |
CN115661107A (en) | Image analysis method, system and equipment based on bladder cancer risk stratification | |
Alksas et al. | A novel computer-aided diagnostic system for early assessment of hepatocellular carcinoma | |
CN114549463A (en) | Curative effect prediction method, system, equipment and medium for breast cancer liver metastasis anti-HER-2 treatment | |
CN113393938B (en) | Breast cancer risk prediction system fusing image and clinical characteristic information | |
CN113889261A (en) | Pathological feature assistance-based PET/CT automatic lung cancer diagnosis classification model training method | |
CN110827275B (en) | Liver nuclear magnetic artery image quality grading method based on raspberry pie and deep learning | |
Kore et al. | A bibliometric approach to track research trends in computer-aided early detection of cancer using biomedical imaging techniques | |
CN114511564B (en) | Image analysis method for breast cancer residual tumor load based on DCE-MRI | |
CN110163195A (en) | Liver cancer divides group's prediction model, its forecasting system and liver cancer to divide group's judgment method | |
Benbakreti et al. | Impact of the data augmentation on the detection of brain tumor from MRI images based on CNN and pretrained models | |
Chang et al. | CT manifestations of gallbladder carcinoma based on neural network | |
CN108346471A (en) | A kind of analysis method and device of pathological data | |
CN111265234A (en) | Method and system for judging properties of lung mediastinal lymph nodes | |
Che | Improved nonalcoholic fatty liver disease diagnosis from ultrasound data based on deep learning |
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: 20181106 |
|
RJ01 | Rejection of invention patent application after publication |