CN109145838B - Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance - Google Patents
Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance Download PDFInfo
- Publication number
- CN109145838B CN109145838B CN201810992268.6A CN201810992268A CN109145838B CN 109145838 B CN109145838 B CN 109145838B CN 201810992268 A CN201810992268 A CN 201810992268A CN 109145838 B CN109145838 B CN 109145838B
- Authority
- CN
- China
- Prior art keywords
- neural network
- probability
- network model
- image
- formula
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000003745 diagnosis Methods 0.000 title claims abstract description 20
- 208000030808 Clear cell renal carcinoma Diseases 0.000 title claims abstract description 17
- 206010073251 clear cell renal cell carcinoma Diseases 0.000 title claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 15
- 238000003062 neural network model Methods 0.000 claims abstract description 36
- 230000003902 lesion Effects 0.000 claims abstract description 22
- 230000001575 pathological effect Effects 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000009826 distribution Methods 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 4
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims 1
- 201000009030 Carcinoma Diseases 0.000 abstract description 5
- 210000004027 cell Anatomy 0.000 description 8
- 208000008839 Kidney Neoplasms Diseases 0.000 description 7
- 206010028980 Neoplasm Diseases 0.000 description 5
- 206010038389 Renal cancer Diseases 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 201000011510 cancer Diseases 0.000 description 4
- 210000003734 kidney Anatomy 0.000 description 4
- 201000010982 kidney cancer Diseases 0.000 description 4
- 238000002271 resection Methods 0.000 description 4
- 208000009956 adenocarcinoma Diseases 0.000 description 3
- 231100000915 pathological change Toxicity 0.000 description 3
- 230000036285 pathological change Effects 0.000 description 3
- 241001164374 Calyx Species 0.000 description 2
- 208000006265 Renal cell carcinoma Diseases 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 208000009060 clear cell adenocarcinoma Diseases 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 206010020524 Hydronephrosis Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000740 bleeding effect Effects 0.000 description 1
- 230000002308 calcification Effects 0.000 description 1
- 239000002775 capsule Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 201000007574 granular cell carcinoma Diseases 0.000 description 1
- 201000011066 hemangioma Diseases 0.000 description 1
- 210000001165 lymph node Anatomy 0.000 description 1
- 230000003211 malignant effect Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000017074 necrotic cell death Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 201000010174 renal carcinoma Diseases 0.000 description 1
- 208000014212 sarcomatoid carcinoma Diseases 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 210000004926 tubular epithelial cell Anatomy 0.000 description 1
- 210000005239 tubule Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/695—Preprocessing, e.g. image segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a renal clear cell carcinoma diagnosis method based on random Gaussian field neural network assistance, which comprises the steps of constructing a data-driven self-learning complex network structure model in a training stage, taking an original medical image and a corresponding marked pathological block as the input and the output of a network, and performing supervised learning on the network; in the decision stage, an unknown input original medical image is given, the probability geometric mean value of the lesion of the part to be detected is calculated, and a confidence interval is given, so that a doctor is assisted to determine the canceration grade and the physiological lesion area. According to the invention, after a data-driven random Gaussian field neural network model is constructed and supervised learning is carried out on the network model, unknown medical images can be effectively discriminated, and a doctor is assisted in positioning a lesion part, so that the doctor is helped to better judge the lesion part and the lesion grade of the renal clear carcinoma cell, the pressure of the doctor is relieved, and the diagnosis accuracy is improved.
Description
Technical Field
The invention relates to the field of medical assistance, in particular to a renal clear cell carcinoma diagnosis method based on random Gaussian field neural network assistance.
Background
Renal parenchymal carcinoma is an adenocarcinoma derived from renal tubular epithelial cells, 85% of which are clear cell carcinomas, and some of which are granular cell carcinomas and mixed cell carcinomas. Cancer often has bleeding, necrosis, cystosis and calcification. It grows in the kidney parenchyma, infiltrates, presses, destroys the renal calyx and calyx of the kidney after growing up, develops outside the renal capsule, and forms hemangioma emboli or transfers to lymph nodes and other organs. Pathologically, kidney cancers are classified as type 4: clear cell type kidney cancer, granular cell type kidney cancer, mixed cell type kidney cancer, undifferentiated cell type kidney cancer. Among them, the majority of clear cell renal carcinoma accounts for 70% -80% of renal carcinoma, and its cancer cells are often arranged in the form of sheet, cord, alveolar or tubular, much like renal tubules. Although clear cell carcinoma is the least malignant of kidney cancers, it is often mixed with granular cell carcinoma and spindle cell carcinoma in clinical practice, and microscopic classification is very difficult. Therefore, a method for fusing the current artificial intelligence leading-edge algorithm is urgently needed, so that the medical contrast images are classified and identified, and a doctor is assisted in diagnosis.
In the field of artificial intelligence signal processing, artificial neural networks are widely used in data classification and parameter mapping due to their capabilities of nonlinear modeling and adaptive data. The artificial neural network relies on the link property between the internal adaptive basis functions to learn and characterize strong correlation between data, reflecting the classification and regression characteristics of the data. On the other hand, a probabilistic stochastic model based on a bayesian framework, such as a stochastic gaussian field, provides a data processing means differentiated from a neural network: the random Gaussian field energy probability prediction result not only provides a point estimation value, but also can analyze a prediction confidence interval, and the method is very helpful to the practical problem. The invention utilizes the advantages of the two to construct a brand new system framework and utilizes the new mathematical model to assist the doctor in diagnosing the medical image.
In the prior art, patients who need renal clear cell carcinoma resection operation after diagnosis usually need to resect the whole kidney of the lesion, and although the idea is simple, the cost for the patients is huge. In the mainstream laparoscopic renal tumor resection operation case at present, a doctor needs to subjectively perform two-dimensional identification on a scanning image of a cancerous physiological tissue block so as to make manual judgment. This procedure often requires a significant amount of clinical experience and knowledge accumulation by the physician to make an effective pre-evaluation. Therefore, the method is time-consuming and labor-consuming, and the effect is not necessarily ideal. To this end, the present invention seeks to improve physician assessment of lesion site location and lesion grade by devising an artificial intelligence assistance mechanism to allow targeted resection rather than total kidney resection.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a renal clear cell carcinoma diagnosis method based on random Gaussian field neural network assistance, which implements intelligent supervised learning by constructing a data-driven artificial intelligent random Gaussian field neural network model and providing a large number of original medical images and identification results as system input and output, and can effectively screen unknown medical images and assist doctors in positioning lesion parts after learning training is completed, so that the doctors are helped to better judge the lesion parts and lesion grades of renal clear carcinoma cells, the pressure of the doctors is relieved, and the diagnosis accuracy is improved.
In order to achieve the purpose, the invention adopts the technical scheme that:
a renal clear cell carcinoma diagnosis method based on random Gaussian field neural network assistance is characterized in that in a training stage, a data-driven self-learning complex network structure model is constructed, an original medical image and a corresponding pathological block with a label are used as input and output of a network, and supervised learning is carried out on the network; in the decision stage, an unknown input original medical image is given, the probability geometric mean value of the lesion of the part to be detected is calculated, and a confidence interval is given, so that a doctor is assisted to determine the canceration grade and the physiological lesion area.
Preferably, the method specifically comprises the following steps:
s01, image preprocessing: converting the original medical image into a gray image with gray values;
s02, constructing and training a neural network model: mapping a gray value of a gray image as input into an intermediate parameter through a random Gaussian field, and linking the gray value to an output layer in a full-connected undirected graph mode, wherein the output layer is set as an identification result through manual labeling, so that a random Gaussian field neural network model with a nonlinear classification function is formed, and after the neural network model is constructed, a large number of original medical images and corresponding identification results with pathological block labels are used as input and output of the neural network model, so that supervised learning is performed on the neural network model;
s03, using the neural network model to identify: after image preprocessing is carried out on the medical image of the part to be detected, the medical image is input into the neural network model, a classification mode is predicted and judged, each classification result has a certain probability, a confidence interval of prediction estimation is obtained at the same time, and a doctor assists diagnosis by transversely comparing the probability of the classification result with the confidence interval.
Further, the original medical image is a medical scan image of medical renal clear cell carcinoma.
Preferably, S01 further includes the steps of: performing one or more of the following operations on the original image by using an image enhancement algorithm: rotation, alignment, uniform image size, enhanced brightness and contrast.
Preferably, S02 further includes the steps of:
(a1) and (3) setting x to represent a vector formed by pixel data of an input image, and y to represent an output category, and constructing a random Gaussian field neural network model as follows:
y(x)=W(x)[f(x)+σf∈]+σyz (1)
wherein,
e ∈ (x) and z ∈ (z) (x) are two white gaussian noise processes with different covariance parameters, respectively, with a probability distribution of e N (0, I)q),N(0,Iq) Denotes a mean of 0, covariance matrix IqIs a Gaussian distribution of a unit matrix of dimension Q x Q, and the probability distribution of z is N (0, I)p),N(0,Ip) Denotes a mean of 0, covariance matrix IpIs a Gaussian distribution, σ, of a P dimensional unit matrixfAnd σyRespectively, the energy coefficient to be estimated,
w (x) is a P × Q matrix, where each element W (x)ijAre all an independent random Gaussian field, i.e.Wherein k iswMay be any form of semi-positive kernel function,
f(x)=f(f1(x),f2(x),…,fq(x) Is a Q-dimensional vector in which any one element is an independent random Gaussian field, i.e.
(a2) Let the training data set beThe unknown model parameters u ═ f, W, f and W represent the incoming data points, respectivelyF (x) and W (x) later, according to the definition of the random Gaussian field, the following prior probability distribution is provided:
p(u|σf,θf,θw)=N(0,CB) (2)
wherein theta isf,θwRespectively, the hyper-parameters contained in the kernel functions referred to in f (x) and W (x), CBIs a block diagonal matrix with NQ (P +1) multiplied by NQ (P +1) dimensions,
while the likelihood function is derived from equation (1) as:
the Bayesian theorem is applied to obtain:
wherein, the formula (4) is a target formula to be estimated, and the estimation of the formula (4) is obtained, namely the estimation of y (x) is obtained to obtain a judgment result;
(a3) and (4) optimizing the formula by using a variational Bayes method so as to obtain the optimal model structure parameters. The essence of variational bayes is to approximate the probability distribution q (-) to be estimated to the true posterior probability distribution p (-) in an iterative manner (i.e., equation 4), i.e., by minimizing the distortion function Dist:
Dist=-H[q(v)]-∫q(v)logp(v)dv (5)
wherein,H[·]representing an entropy function, first of allAssigning an inverse gamma distribution (IG), namely:
second, the approximate distribution q (v) is designed as follows:
whereinAre all in inverse gamma distribution,are all in a Gaussian distribution with an N dimension,
and finally, carrying out iteration to obtain an optimal value for the formula (7), namely carrying out piecewise linear search in the gradient direction by using a conjugate gradient descent method to find out theta for maximizing the formula (7)f,θw。
Preferably, S03 further includes the steps of:
(b1) for a new unknown medical image, firstly calling the step S01 to carry out image preprocessing;
(b2) using the network model trained in the step S02 to perform classification decision, firstly according to the bayesian formula, the accurate target prediction function should be:
then, approximation is performed by the thought of the variational Bayes method, and the above formula is setThat is, the standard probability distribution can be obtained by multiplying two approximate probability distributions, and the above formula is integrated according to the attributes of the conditional probability and the marginal probability of the standard Gaussian distribution to obtain the mean (y) of the estimation decision*) Sum variance cov (y)*)ijVariance cov (y)*)ijI.e. the confidence interval, as follows:
wherein k represents the discrimination type, δijThe method is characterized in that the method is a kronecker delta function, on the basis of obtaining a decision mean value, the type with the maximum probability is selected as a decision result through transverse comparison, and meanwhile, the probability of variation of the decision result is determined according to a variance function, so that a complete judgment standard is formed.
Preferably, the random Gaussian field parameters and the neural network model parameters are estimated by a variational Bayes method under a marginalized maximum likelihood criterion.
Compared with the prior art, the method has the beneficial effects that 1) a neural network model based on a random Gaussian field is constructed, a large number of original medical images and corresponding marked pathological blocks are used as input and output of the neural network model, supervised learning is carried out on the neural network model, then unknown new medical images of a part to be detected are input to the constructed neural network model which is trained by the supervised learning, the mean value and confidence interval of probability estimation and judgment of the pathological changes of the part to be detected are obtained, a doctor is assisted to determine the cancer grade and the physiological pathological blocks, and the positioning of the doctor on the pathological changes and the accuracy of evaluation of the pathological changes are improved; 2) the method is used for auxiliary diagnosis and treatment of medical images, integrates an artificial intelligence algorithm, and can help doctors to better judge the lesion part and the lesion grade of the renal clear cancer cells, so that the pressure of the doctors is relieved, the diagnosis accuracy is improved, and the working efficiency is improved; 3) the neural network model disclosed by the invention is based on a stochastic Gaussian field, not only contains probability characteristics, but also contains non-frequency characteristics of the neural network, can be used for carrying out non-linear probability classification on unknown images, and provides a confidence interval analysis result, so that the reliability of auxiliary diagnosis is greatly improved.
Drawings
Fig. 1 is a flow diagram of a stochastic gaussian field neural network-based assisted renal clear cell carcinoma diagnostic method according to an embodiment;
FIG. 2 is a schematic diagram of a stochastic Gaussian field neural network model of the present invention;
FIG. 3 is a schematic illustration of a renal clear cell carcinoma lesion of the present invention according to an embodiment;
fig. 4 is a schematic view of a renal clear cell carcinoma lesion of the present invention, according to an embodiment.
Detailed Description
The present invention will be described in further detail with reference to the following embodiments, which are illustrative only and not limiting, and the scope of the present invention is not limited thereby.
The invention provides a renal clear cell carcinoma diagnosis method based on random Gaussian field neural network assistance, which comprises the steps of constructing a data-driven self-learning complex network structure model in a training stage, taking an original medical image and a corresponding pathological block with a label as the input and the output of a network, and performing supervised learning on the network; in the decision stage, an unknown input original medical image is given, the probability geometric mean value of the lesion of the part to be detected is calculated, and a confidence interval is given, so that a doctor is assisted to determine the canceration grade and the physiological lesion area. As shown in the attached drawings, the method specifically comprises the following steps:
s01, image preprocessing: converting an original medical image into a gray image with gray value representation, and performing operations such as rotation, alignment, image size unification, brightness enhancement, contrast enhancement and the like by using a traditional image enhancement algorithm, wherein the original medical image is a medical scanning image of medical renal clear cell carcinoma, and the marked pathological blocks comprise the pathological blocks marked with calculi and hydronephrosis and the pathological blocks marked with tumors;
s02, constructing and training a neural network model: mapping a gray value of a gray image as input into an intermediate parameter through a random Gaussian field, and linking the gray value to an output layer in a full-connected undirected graph mode, wherein the output layer is set as an identification result through manual labeling, so that a random Gaussian field neural network model with a nonlinear classification function is formed, and after the neural network model is constructed, a large number of original medical images and corresponding identification results with pathological block labels are used as input and output of the neural network model, so that supervised learning is performed on the neural network model; wherein, the random Gaussian field parameters and the neural network model parameters are obtained by estimation by a variational Bayes method under a marginalized maximum likelihood criterion;
s03, using the neural network model to identify: after the medical image of the part to be detected is subjected to image preprocessing, the medical image is input into the neural network model, a classification mode is predicted and judged, each classification result has a certain probability, a confidence interval of prediction estimation is obtained at the same time, then the classification result is predicted and judged through the neural network model, each classification result has a certain probability, the system provides the confidence interval of the prediction judgment result, and a doctor can perform auxiliary diagnosis by transversely comparing the probability of the classification result with the probability coverage range (confidence interval) of possibility, so that the canceration grade and the physiological lesion block of the part to be detected are determined.
Specifically, as shown in fig. 2, S02 further includes the following steps:
(a1) and (3) setting x to represent a vector formed by pixel data of an input image, and y to represent an output category, and constructing a random Gaussian field neural network model as follows:
y(x)=W(x)[f(x)+σf∈]+σyz (1)
wherein,
e ∈ (x) and z ∈ (z) (x) are two white gaussian noise processes with different covariance parameters, respectively, with a probability distribution of e N (0, I)q),N(0,Iq) Denotes a mean of 0, covariance matrix IqIs a Gaussian distribution of a unit matrix of dimension Q x Q, and the probability distribution of z is N (0, I)p),N(0,Ip) Denotes a mean of 0, covariance matrix IpIs a Gaussian distribution, σ, of a P dimensional unit matrixfAnd σyRespectively, the energy coefficient to be estimated,
w (x) is a P × Q matrix, where each element W (x)ijAre all an independent random Gaussian field, i.e.Wherein k iswMay be any form of semi-positive kernel function,
f(x)=f(f1(x),f2(x),…,fq(x) Is a Q-dimensional vector in which any one element is an independent random Gaussian field, i.e.
(a2) Let the training data set beThe unknown model parameters u ═ f, W, f and W represent the incoming data points, respectivelyF (x) and W (x) later, according to the definition of the random Gaussian field, the following prior probability distribution is provided:
p(u|σf,θf,θw)=N(0,CB) (2)
wherein theta isf,θwRespectively, the hyper-parameters contained in the kernel functions referred to in f (x) and W (x), CBIs a block diagonal matrix with NQ (P +1) multiplied by NQ (P +1) dimensions,
while the likelihood function is derived from equation (1) as:
the Bayesian theorem is applied to obtain:
wherein, the formula (4) is a target formula to be estimated, and the estimation of the formula (4) is obtained, namely the estimation of y (x) is obtained to obtain a judgment result;
(a3) and (4) optimizing the formula by using a variational Bayes method so as to obtain the optimal model structure parameters. The essence of variational bayes is to approximate the probability distribution q (-) to be estimated to the true posterior probability distribution p (-) in an iterative manner (i.e., equation 4), i.e., by minimizing the distortion function:
Dist=-H[q(v)]-∫q(v)logp(v)dv (5)
wherein,H[·]representing an entropy function, first of allAssigning an inverse gamma distribution (IG), namely:
wherein the energy coefficient sigma to be estimatedfAnd σyOne can derive from equation (6), and next, design the approximate distribution q (v) as follows:
whereinAre all in inverse gamma distribution,are all in a Gaussian distribution with an N dimension,
and finally, carrying out iteration to obtain an optimal value for the formula (7), namely carrying out piecewise linear search in the gradient direction by using a conjugate gradient descent method to find out theta for maximizing the formula (7)f,θw。
Specifically, S03 further includes the steps of:
(b1) for a new unknown medical image, firstly calling the step S01 to carry out image preprocessing;
(b2) using the network model trained in the step S02 to perform classification decision, firstly according to the bayesian formula, the accurate target prediction function should be:
then, approximation is performed by the thought of the variational Bayes method, and the above formula is setThat is, the standard probability distribution can be obtained by multiplying two approximate probability distributions, and the above formula is integrated according to the attributes of the conditional probability and the marginal probability of the standard Gaussian distribution to obtain the mean (y) of the estimation decision*) Sum variance cov (y)*)ijVariance cov (y)*)ijI.e. the confidence interval, as follows:
wherein k represents the discrimination type, δijThe method is characterized in that the method is a kronecker delta function, on the basis of obtaining a decision mean value, the type with the maximum probability is selected as a decision result through transverse comparison, and meanwhile, the probability of variation of the decision result is determined according to a confidence interval, so that a complete judgment standard is formed.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention.
Claims (3)
1. A renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance is characterized in that,
in the training stage, a data-driven self-learning complex network structure model is constructed, the original medical image and the corresponding marked pathological block are used as the input and the output of the network, and the network is subjected to supervised learning; in the judging stage, an unknown input original medical image is given, the probability geometric mean value of the lesion of the part to be detected is calculated, and a confidence interval is given, so that a doctor is assisted to determine the canceration grade and the physiological lesion area; the method specifically comprises the following steps:
s01, image preprocessing: converting the original medical image into a gray image with gray values;
s02, constructing and training a neural network model: mapping a gray value of a gray image as input into an intermediate parameter through a random Gaussian field, and linking the gray value to an output layer in a full-connected undirected graph mode, wherein the output layer is set as an identification result through manual labeling, so that a random Gaussian field neural network model with a nonlinear classification function is formed, and after the neural network model is constructed, a large number of original medical images and corresponding identification results with pathological block labels are used as input and output of the neural network model, so that supervised learning is performed on the neural network model;
s03, using the neural network model to identify: after image preprocessing is carried out on a medical image of a part to be detected, the medical image is input into the neural network model, a classification mode is predicted and judged, each classification result has a certain probability, a confidence interval of prediction estimation is obtained at the same time, and a doctor assists diagnosis by transversely comparing the probability of the classification result with the confidence interval;
s02 further includes the steps of:
(a1) and (3) setting x to represent a vector formed by pixel data of an input image, and y to represent an output category, and constructing a random Gaussian field neural network model as follows:
y(x)=W(x)[f(x)+σf∈]+σyz (1)
wherein,
e ∈ (x) and z ∈ (z) (x) are two white gaussian noise processes with different covariance parameters, respectively, with a probability distribution of e N (0, I)q),N(0,Iq) Denotes a mean of 0, covariance matrix IqIs a Gaussian distribution of a unit matrix of dimension Q x Q, and the probability distribution of z is N (0, I)p),N(0,Ip) Denotes a mean of 0, covariance matrix IpIs a Gaussian distribution, σ, of a P dimensional unit matrixfAnd σyRespectively, the energy coefficient to be estimated,
w (x) is a P × Q matrix, where each element W (x)ijAre all an independent random Gaussian field, i.e.Wherein k iswIs a semi-positive kernel function of any form,
f(x)=f(f1(x),f2(x),…,fq(x) Is a Q-dimensional vector in which any one element is an independent random Gaussian field, i.e.
(a2) Let the training data set beUnknown model parametersf and W represent the carry-in data points, respectivelyF (x) and W (x) later, according to the definition of the random Gaussian field, the following prior probability distribution is provided:
wherein theta isf,θwRespectively, the hyper-parameters contained in the kernel functions referred to in f (x) and W (x), CBIs a block diagonal matrix with NQ (P +1) multiplied by NQ (P +1) dimensions,
and simultaneously obtaining a likelihood function according to the formula (1) as follows:
and (3) obtaining by applying Bayesian theorem:
wherein, the formula (4) is a target formula to be estimated, and the estimation of the formula (4) is obtained, namely the estimation of y (x) is obtained to obtain a judgment result;
(a3) and (4) optimizing the formula (4) by using a variational Bayes method to obtain the optimal model structure parameters, wherein the essence of the variational Bayes method is that the probability distribution q (-) to be estimated approaches the real posterior probability distribution p (-) in an iterative mode, namely the formula (4), namely by minimizing the distortion function Dist:
Dist=-H[q(v)]-∫q(v)logp(v)dv (5)
wherein,H[·]representing an entropy function, first of allAssigning an inverse gamma distribution (IG), namely:
second, the approximate distribution q (v) is designed as follows:
whereinAre all in inverse gamma distribution,are all in a Gaussian distribution with an N dimension,
and finally, carrying out iteration to obtain an optimal value for the formula (7), namely carrying out piecewise linear search in the gradient direction by using a conjugate gradient descent method to find out theta for maximizing the formula (7)f,θw;
S03 further includes the steps of:
(b1) for a new unknown medical image, firstly calling the step S01 to carry out image preprocessing;
(b2) using the network model trained in the step S02 to perform classification decision, firstly according to the bayesian formula, the accurate target prediction function should be:
then, approximation is performed by the thought of the variational Bayes method, and the above formula is setThat is, the standard probability distribution can be obtained by multiplying two approximate probability distributions, and the above formula is integrated according to the attributes of the conditional probability and the marginal probability of the standard Gaussian distribution to obtain the mean (y) of the estimation decision*) Sum variance cov (y)*)ijVariance cov (y)*)ijI.e. confidence levelIntervals, as follows:
wherein k represents the discrimination type, δijThe method is characterized in that the method is a kronecker delta function, on the basis of obtaining a decision mean value, the type with the maximum probability is selected as a decision result through transverse comparison, and meanwhile, the result is determined to have large variation possibility according to a variance function, so that a complete judgment standard is formed.
2. The stochastic gaussian field neural network-assisted renal clear cell carcinoma diagnosis method according to claim 1, wherein S01 further comprises the steps of: performing one or more of the following operations on the original image by using an image enhancement algorithm: rotation, alignment, uniform image size, enhanced brightness and contrast.
3. The stochastic gaussian field neural network-assisted renal clear cell carcinoma-based diagnosis method according to claim 1, wherein the stochastic gaussian field parameters and the neural network model parameters are estimated by a variational bayes method under a marginalized maximum likelihood criterion.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810992268.6A CN109145838B (en) | 2018-08-29 | 2018-08-29 | Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810992268.6A CN109145838B (en) | 2018-08-29 | 2018-08-29 | Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109145838A CN109145838A (en) | 2019-01-04 |
CN109145838B true CN109145838B (en) | 2021-08-17 |
Family
ID=64828713
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810992268.6A Active CN109145838B (en) | 2018-08-29 | 2018-08-29 | Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109145838B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109820507B (en) * | 2019-03-25 | 2024-03-08 | 钟霁媛 | Disease auxiliary diagnosis method and device |
CN110444294B (en) * | 2019-09-10 | 2023-04-25 | 中南大学 | Auxiliary analysis method and equipment for prostate cancer based on perception neural network |
CN111553505A (en) * | 2019-11-26 | 2020-08-18 | 国网浙江省电力有限公司 | Method for predicting warehousing quantity of electric power materials |
CN111653356A (en) * | 2020-04-20 | 2020-09-11 | 浙江大学 | New coronary pneumonia screening method and new coronary pneumonia screening system based on deep learning |
CN112242008B (en) * | 2020-10-17 | 2021-09-07 | 郑州西亚斯学院 | AR interaction visual perception method and system for transparent container |
CN112508884B (en) * | 2020-11-24 | 2024-09-27 | 江苏大学 | Comprehensive detection device and method for cancerous region |
CN112464579B (en) * | 2021-02-02 | 2021-06-01 | 四川大学 | Identification modeling method for searching esophageal cancer lesion area based on evolutionary neural network structure |
CN112951427B (en) * | 2021-03-16 | 2023-12-08 | 黑龙江机智通智能科技有限公司 | System for classifying abnormal cells |
CN116269749B (en) * | 2023-03-06 | 2023-10-10 | 东莞市东部中心医院 | Laparoscopic bladder cancer surgical system with improved reserved nerves |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106874687A (en) * | 2017-03-03 | 2017-06-20 | 深圳大学 | Pathological section image intelligent sorting technique and device |
WO2017156640A1 (en) * | 2016-03-18 | 2017-09-21 | Fluent.Ai Inc. | Method and device for automatically learning relevance of words in a speech recognition system |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11056243B2 (en) * | 2015-12-21 | 2021-07-06 | Elekta Ab (Publ) | Systems and methods for optimizing treatment planning |
-
2018
- 2018-08-29 CN CN201810992268.6A patent/CN109145838B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017156640A1 (en) * | 2016-03-18 | 2017-09-21 | Fluent.Ai Inc. | Method and device for automatically learning relevance of words in a speech recognition system |
CN106874687A (en) * | 2017-03-03 | 2017-06-20 | 深圳大学 | Pathological section image intelligent sorting technique and device |
Non-Patent Citations (3)
Title |
---|
Adversarial Phenomenon in the Eyes of Bayesian Deep Learning;Ambrish Rawat et.al;《arXiv:1711.08244v1 [stat.ML]》;20171122;第1-13页 * |
Bayesian Deep Convolutional Encoder-Decoder Networks for Surrogate Modeling and Uncertainty Quantification;Yinhao Zhu et.al;《arXiv:1801.06879v1 [physics.comp-ph]》;20180121;第1-52页 * |
虚幻面孔加工有效连通网络的非线性DCM分析;李军 等;《西安电子科技大学学报(自然科学版)》;20110630;第38卷(第3期);第69-75页 * |
Also Published As
Publication number | Publication date |
---|---|
CN109145838A (en) | 2019-01-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109145838B (en) | Renal clear cell carcinoma diagnosis method based on stochastic Gaussian field neural network assistance | |
Gunasekara et al. | A systematic approach for MRI brain tumor localization and segmentation using deep learning and active contouring | |
WO2018120942A1 (en) | System and method for automatically detecting lesions in medical image by means of multi-model fusion | |
US10055551B2 (en) | Systems and methods for quantitative analysis of histopathology images using multiclassifier ensemble schemes | |
Dahab et al. | Automated brain tumor detection and identification using image processing and probabilistic neural network techniques | |
US20170169276A1 (en) | Systems and methods for automated screening and prognosis of cancer from whole-slide biopsy images | |
Patel et al. | GARL-Net: Graph based adaptive regularized learning deep network for breast cancer classification | |
Hamad et al. | Breast cancer detection and classification using artificial neural networks | |
Wang et al. | Exploring automatic prostate histopathology image gleason grading via local structure modeling | |
Mesleh | Lung Cancer Detection Using Multi-Layer Neural Networks with Independent Component Analysis: A Comparative Study of Training Algorithms. | |
CN116740435A (en) | Breast cancer ultrasonic image classifying method based on multi-mode deep learning image group science | |
Akkar et al. | Diagnosis of lung cancer disease based on back-propagation artificial neural network algorithm | |
CN117994238A (en) | Thoracic surgery image analysis method based on big data | |
Kumar et al. | A novel glaucoma detection model using Unet++-based segmentation and ResNet with GRU-based optimized deep learning | |
Shen et al. | Identify representative samples by conditional random field of cancer histology images | |
CN111445456B (en) | Classification model, training method and device of network model, and recognition method and device | |
Tasoulis et al. | Urinary bladder tumor grade diagnosis using on-line trained neural networks | |
Vijila Rani et al. | Detection of cervix tumor using an intelligent system accompanied with PNN classification approach | |
Niwas et al. | Complex wavelet as nucleus descriptors for automated cancer cytology classifier system using ANN | |
Pamula Pullaiah et al. | Detection of Breast Cancer on Magnetic Resonance Imaging Using Hybrid Feature Extraction and Deep Neural Network Techniques. | |
Khoulqi et al. | Cervical Cancer Detection and Classification Using MRIs | |
Huang et al. | Semantic segmentation of breast ultrasound image with pyramid fuzzy uncertainty reduction and direction connectedness feature | |
Govinda et al. | Artificial neural networks in UWB image processing for early detection of breast cancer | |
Yan et al. | A Novel Adaptive Fuzzy Deep Learning Approach for Histopathologic Cancer Detection | |
Duraisamy et al. | CNN-based approach for segmentation of brain and lung MRI 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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |