CN114783593A - Method and system for automatically detecting kidney diseases based on machine learning - Google Patents

Method and system for automatically detecting kidney diseases based on machine learning Download PDF

Info

Publication number
CN114783593A
CN114783593A CN202210514808.6A CN202210514808A CN114783593A CN 114783593 A CN114783593 A CN 114783593A CN 202210514808 A CN202210514808 A CN 202210514808A CN 114783593 A CN114783593 A CN 114783593A
Authority
CN
China
Prior art keywords
kidney
neural network
particle
iteration
image
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
Application number
CN202210514808.6A
Other languages
Chinese (zh)
Inventor
罗学敏
王洪平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Deao Smart Medical Technology Co ltd
Original Assignee
Individual
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN202210514808.6A priority Critical patent/CN114783593A/en
Publication of CN114783593A publication Critical patent/CN114783593A/en
Pending legal-status Critical Current

Links

Images

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
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30084—Kidney; Renal

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Public Health (AREA)
  • Radiology & Medical Imaging (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Image Analysis (AREA)

Abstract

Machine learning-based kidney disease automatic detection method and system, including kidney image collection module, kidney image management module, kidney image intelligent diagnosis module and kidney disease classification module, the kidney image that has the kidney disease diagnosis label of collecting through kidney image collection module, store and update the kidney image that the kidney disease type corresponds at kidney image management module, utilize machine learning algorithm to infer the kidney disease type through kidney image intelligent diagnosis module, kidney disease classification module classifies the kidney disease of diagnosis according to different kidney disease types, thereby realize the high-efficient, accurate diagnosis of each kidney disease type, and it is significant to the diagnosis and the early screening of kidney disease.

Description

Method and system for automatically detecting kidney diseases based on machine learning
Technical Field
The present disclosure relates to systems and methods for image analysis and medical diagnostic testing using image analysis, and more particularly, to a method and system for machine learning-based automated detection of kidney disease.
Background
Medical imaging refers to the technique and process of obtaining internal images of a human body or a part of a human body in a non-invasive manner for medical treatment or medical research. With the rapid development of various scientific technologies, medical image processing technology has also made rapid progress, but with the development of scientific technologies and the popularization of medical image applications, more and more medical images need to be interpreted by doctors, and medical image interpretation gradually becomes a challenging task. The traditional medical image interpretation and diagnosis process mainly depends on experienced doctors, the process has subjectivity, in addition, the phenomenon of interpretation error caused by the cognitive ability limitation or fatigue of doctors can also occur in the manual medical image interpretation process, so that misdiagnosis is caused, and the defects indicate the importance of using the effective medical image analysis technology to improve the accuracy of disease diagnosis results.
The artificial intelligence technology is composed of different fields such as machine learning and computer vision, and aims to produce an intelligent machine similar to human beings. According to the problems existing in medical image interpretation, researchers apply image processing technology and artificial intelligence technology to medical image diagnosis, so that the automatic analysis of medical images and the function of assisting doctors in making medical diagnosis are achieved, and the workload of the doctors can be effectively reduced.
Disclosure of Invention
In view of the above problems, the present invention aims to provide a method and a system for automatic detection of kidney diseases based on machine learning.
The purpose of the invention is realized by the following technical scheme:
the kidney disease automatic detection method and system based on machine learning comprise a kidney image collection module, a kidney image management module, a kidney influence intelligent diagnosis module and a kidney disease classification module;
kidney image collection module: the system is used for collecting kidney disease images of various types of patients, wherein the sources of the kidney images comprise InterVar (site pathogenicity judgment), GeneReviews (disease database) and kidney images of patients in a hospital;
kidney image management module: the kidney image pre-storing unit is used for pre-storing various types of kidney images collected by the kidney image collecting module, screening the kidney images with kidney disease diagnosis labels through the kidney image updating unit, cleaning excessive noise, default or irrelevant images, and then updating a kidney image training set;
the intelligent kidney image diagnosis module comprises a kidney image processing unit and a kidney image database, wherein the kidney image processing unit extracts the disease characteristics of a kidney image by analyzing the kidney image with a kidney disease diagnosis label, a machine learning model is established by taking the kidney image updated by a kidney image updating unit as a training set, the kidney image to be judged is automatically detected for the disease, the trained kidney image with the kidney disease diagnosis label is stored in the kidney image database, and a distinguishable kidney image automatic disease detection model is generated so as to be convenient for next time without retraining and be directly used;
the kidney disease classification module further classifies the kidney diseases judged by the kidney image intelligent diagnosis module so as to ensure that a patient using the system can obtain a diagnosis and treatment scheme through self-checking kidney images.
Further, the kidney image collection module collects images from sources of InterVar (site pathogenicity assessment), GeneReviews (disease database) and kidney images of in-hospital patients, and collects the images to the sql database for online management.
Further, the kidney image updating unit screens the kidney images with the diagnosis labels of the kidney diseases, wherein the labels comprise calcific foci, density of renal pelvis of renal calyx, renal medulla, cyst and cancer embolus.
Furthermore, the kidney image processing unit performs image segmentation on the kidney image to be diagnosed by adopting a maximum entropy multi-threshold segmentation method, and determines an optimal combination threshold of the maximum entropy multi-threshold segmentation method adopted in the kidney image processing unit by adopting a particle swarm optimization.
Further, in the particle swarm algorithm, the particle swarm iteratively updates the solution of the particle in the following manner:
Vi(t+1)＝ωVi(t)+ci1(t)r1(Pbesti(t）-Xi(t))+ci2(t)r2(Gbest(t)-Xi(t))
Xi(t+1)＝Xi(t)+Vi(t+1)
in the formula, Xi(t +1) and Vi(t +1) denotes the solution and velocity of particle i at the (t +1) th iteration, X, respectivelyi(t) and Vi(t) respectively represents the solution and velocity of the particle i at the t-th iteration, ω represents an inertial weight factor, Pbesti(t) represents the historical optimal solution of particle i at the t-th iteration, Gbest (t) represents the global optimal solution of the particle swarm at the t-th iteration, r1And r2Representing a randomly generated random number between 0 and 1, ci1(t) cognitive learning factor for particle i at the t-th iteration, ci2(t) represents the social learning factor of particle i at the t-th iteration.
Further, setting a particle swarm to adjust the cognitive learning factor c of the particle i at the t-th iteration in the following wayi1(t) and social learning factor ci2Value of (t):
let omegai(t) represents the local neighborhood of particle i at the t-th iteration, and Ωi(t) is to solve Xi(t) a circular region with R (t) as the center and R (t) as the radius, R (t) is the local neighborhood radius of the particle swarm in the t iteration, and the value of R (t) is set as follows:
Figure BDA0003641056930000031
wherein R isi(t) represents the local neighborhood radius of particle i at the t-th iteration, and
Figure BDA0003641056930000032
Figure BDA0003641056930000033
denotes the dissociation X of the particle population at the t-th iterationi(t) a solution for the first nearest particle, c representing a given positive integer, and c < N, N representing the number of particles in the population; omegai,best(t) represents the historical optimum neighborhood of particle i at the t-th iteration, and Ωi,best(t) is to solve Pbesti(t) a circular area with r (t) as the center and r (t) as the radius, r (t) is the historical optimal neighborhood radius of the particle swarm in the t iteration, and
Figure BDA0003641056930000034
wherein r isi(t) represents the historical optimal neighborhood radius of particle i at the t-th iteration, an
Figure BDA0003641056930000035
In the formula,
Figure BDA0003641056930000036
denotes the dissociation of Pbest in the population at the t-th iterationi(t) solution of the a-th particle;
defining a function qi(t) for detecting the optimum symmetry of the particle i in the tth iteration, q is setiThe formula for (t) is:
Figure BDA0003641056930000037
in the formula, qi(t) shows the optimum uniformity, J, of particle i at the tth iterationi(t) shows the local optimum symmetry, S, of particle i at the tth iterationi(t) represents the historical goodness of the particle i at the tth iteration, Ji(t) and SiThe values of (t) are:
Figure BDA0003641056930000038
Figure BDA0003641056930000039
in the formula, Xj(t) represents the solution of particle j at the t-th iteration, Xb(t) represents the solution of particle b at the t-th iteration, and Xj(t)≠Xb(t)，Xk(t) represents the solution of particle k at the t-th iteration, Xo(t) watchShows the solution of particle o at the t-th iteration, and Xk(t)≠Xo(t)，λi,j(t) is a determination function for the optimization between particle i and particle j at the t-th iteration, let fi(t) and fj(t) represents the fitness function values of particle i and particle j at the t-th iteration respectively, then
Figure BDA0003641056930000041
λi,b(t) is a determination function for the optimization between particle i and particle b at the t-th iteration, let fb(t) represents the fitness function value of particle b at the t-th iteration, then
Figure BDA0003641056930000042
mi,best(t) represents the historical optimum neighborhood Ωi,bestThe number of particles in (t);
when q isi(t)＜Hq1(t) or qi(t)＞Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure BDA0003641056930000043
Figure BDA0003641056930000044
when H is presentq1(t)≤qi(t)≤Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure BDA0003641056930000045
Figure BDA0003641056930000046
in the above formula, α and β are two constants for controlling ci1(t) and ci2(t) range of，Hq1(t) is the lower bound of the global optimization symmetry of the particle swarm at the t-th iteration, and
Figure BDA0003641056930000047
Hq2(t) is the upper limit value of the global optimization symmetry of the particle swarm in the t-th iteration, and
Figure BDA0003641056930000048
wherein,
Figure BDA0003641056930000049
means representing the optimum symmetry of the particles in the population at the t-th iteration, an
Figure BDA00036410569300000410
TmaxGiven a maximum number of iterations.
Furthermore, the kidney diseases judged by the intelligent kidney image diagnosis module are further classified by utilizing a back propagation algorithm so as to ensure that a patient using the system can obtain a diagnosis and treatment scheme through the self-checking kidney image, and the method specifically comprises the following steps:
firstly, building an M-layer neural network, and for the output of the (M +1) th-layer network, the following steps are provided:
am+1＝fm+1(Wm+1am+bm+1),m＝0,1,…,M-1
where M is the number of network layers, a is the output of the neural network, i.e., the class of renal disease, there is a0P and aM，am+1Is the output result of the (m +1) th layer neural network, amIs the output result of the mth layer neural network, fm+1Is the excitation function of the neural network of layer m +1, Wm+1Is the weight of the (m +1) th neural network, bm+1Is the bias of the (m +1) th layer neural network, assuming that for each type of input p of the kidney disease characteristics, there is a corresponding kidney disease type output t, the following correspondence can be established:
{p1,t1},{p2,t2},…,{pq,tq},…,{pQ,tQ}
wherein p is1Is the input of the 1 st kidney image feature, t1Is the expected output, p, for the 1 st class2Is the input of the 2 nd kidney image feature, t2Is the expected output, p, for the 2 nd classqIs the input of the qth renal image feature, tqIs the expected output, p, corresponding to the qth classQIs the input of the Q-th renal image feature, tQIs the desired output for the qth class, the algorithm will adjust the network parameters to minimize the mean squared error: f (x) E [ (t-a)2]＝E[(t-a)T(t-a)]Let E-t-a denote the error of the desired output from the predicted output, where F denotes the mean square error, x denotes the arguments of F, which in this case include t and a, E denotes the mathematical expectation, (t-a)TTranspose of the representation error by
Figure BDA0003641056930000051
Instead of f (x), the mean square error is calculated approximately:
Figure BDA0003641056930000052
wherein k is the kth iteration, and the steepest descent method of the approximate mean square error is as follows:
Figure BDA0003641056930000053
Figure BDA0003641056930000054
wherein a is a learning speed of the learning,
Figure BDA0003641056930000055
represents the weight of the ith element to the jth element at the (k +1) th iteration of the mth layer neural network,
Figure BDA0003641056930000056
representing the weight of the ith element to the jth element under the kth iteration of the mth layer neural network, and defining a chain rule as follows:
Figure BDA0003641056930000057
if f (n) is e ^ n and n is 2w, then f (n) (w) is e ^ 2w, have
Figure BDA0003641056930000058
Figure BDA0003641056930000059
Then use the chain rule
Figure BDA00036410569300000510
And
Figure BDA00036410569300000511
can be written as:
Figure BDA00036410569300000512
for the net input n of the mth layer, there are:
Figure BDA00036410569300000513
wherein s represents sensitivity, sm-1Representing the sensitivity of the (m-1) th layer neural network,
Figure BDA00036410569300000514
representing the net input of the ith element in an m-layer neural network,
Figure BDA00036410569300000515
represents the output of the jth element in an m-1 layer neural network,
Figure BDA00036410569300000516
represents the bias of the ith element in an m-layer neural network, thus:
Figure BDA00036410569300000517
if defined:
Figure BDA00036410569300000518
then
Figure BDA00036410569300000519
And
Figure BDA00036410569300000520
can be simplified as follows:
Figure BDA00036410569300000521
expressed as:
Figure BDA0003641056930000061
Figure BDA0003641056930000062
wherein,
Figure BDA0003641056930000063
representing the bias of the ith element in the (k +1) th iteration of the m-layer neural network, the matrix form can be expressed as:
Wn(k+1)＝wm(k)-αsm(am-1)T
bm(k+1)＝bm(k)-αsm
wherein, Wm(k +1) represents wm(k +1) matrix form, i.e. set of weights after k +1 iterations of the m-layer neural network, where smComprises the following steps:
Figure BDA0003641056930000064
wherein,
Figure BDA0003641056930000065
the net input to the mth layer representing the approximation error is partial,
Figure BDA0003641056930000066
1 st net input to m-th layer representing approximation errorThe partial derivatives of the light beams are deflected,
Figure BDA0003641056930000067
the 2 nd net input representing the approximation error to the mth layer is partial-derivative,
Figure BDA0003641056930000068
s-th layer representing approximation error pairmThe net inputs are subjected to partial derivation, and the Jacobian matrix is recorded as:
Figure BDA0003641056930000069
wherein,
Figure BDA00036410569300000610
representing the partial derivative of the net input to the layer m +1 neural network to the net input to the layer m neural network,
Figure BDA00036410569300000611
represents the partial derivative of the 1 st net input of the (m +1) th layer neural network to the 1 st net input of the m' th layer neural network,
Figure BDA00036410569300000612
representing the partial derivative of the 1 st net input of the (m +1) th layer neural network to the 2 nd net input of the m' th layer neural network,
Figure BDA00036410569300000613
represents the s of the 1 st net input of the m +1 st layer neural network to the m layer neural networkmThe partial derivative of the individual net inputs,
Figure BDA0003641056930000071
represents the partial derivative of the 2 nd net input of the m +1 th layer neural network to the 1 st net input of the m layer neural network,
Figure BDA0003641056930000072
representing the net 2 input of the (m +1) th neural network to the net 2 input of the (m) th neural networkThe partial derivatives of the light beams are reflected by the light beam,
Figure BDA0003641056930000073
represents the s < nd > net input of the (m +1) th layer neural network to the m < th > neural networkmThe partial derivative of the individual net inputs,
Figure BDA0003641056930000074
s representing the (m +1) th neural networkm+1The partial derivative of the net input to the net input of the mth layer neural network,
Figure BDA0003641056930000075
s represents the (m +1) th layer neural networkm+1Partial derivatives of the 1 st net input to the mth layer neural network,
Figure BDA0003641056930000076
s representing the (m +1) th neural networkm+1S of net input to mth layer neural networkmThe partial derivative of the individual net inputs,
Figure BDA0003641056930000077
s representing the (m +1) th neural networkm+1S of net input to mth layer neural networkmPartial derivatives of the individual net inputs, taking into account the i, j elements of the matrix:
Figure BDA0003641056930000078
Figure BDA0003641056930000079
wherein:
Figure BDA00036410569300000710
the jacobian matrix can then be written as:
Figure BDA00036410569300000711
wherein:
Figure BDA00036410569300000712
Fm(nm) Represents the net input of the mth layer neural network as nmThe objective function of the time of day,
Figure BDA00036410569300000713
represents a net input to the mth layer neural network of
Figure BDA00036410569300000714
The objective function of the time of day,
Figure BDA00036410569300000715
represents a net input to the mth layer neural network of
Figure BDA00036410569300000716
The objective function of the time of day,
Figure BDA00036410569300000717
represents a net input to the mth layer neural network of
Figure BDA00036410569300000718
And (3) writing a sensitivity recurrence relation formula by a chain rule of a matrix form according to the time target function:
Figure BDA00036410569300000719
Figure BDA00036410569300000720
further, the optimal training weight is obtained by utilizing a back propagation algorithm and a chain method, and the disease information of the patient is input into a trained machine learning model, so that whether the patient has calcified foci, density of renal pelvis, renal medullary, cyst and cancer embolism diseases or not can be automatically detected.
Further, to the machine learning model who establishes, realize the kidney image automated inspection disease to needs are differentiated, will train good kidney image that has kidney disease diagnosis label and keep to kidney image database, generate the kidney image automated inspection disease's that can differentiate model and need not retraining once more in order to make things convenient for next time, directly use, kidney disease classification module is further categorised to the kidney disease that kidney image intelligent diagnosis module differentiateed to guarantee that the patient that uses this system can obtain diagnosis and treatment scheme through self-checking kidney image.
The beneficial effects of the invention are as follows:
(1) the artificial intelligence technology is applied to kidney image analysis, tumor diagnosis models corresponding to human body parts based on kidney images are established, tumor diagnosis can be effectively carried out on the human body parts, and the functions of automatic analysis of the kidney images and medical tumor diagnosis assisting doctors are realized;
(2) the method comprises the steps of performing image segmentation on a kidney image to be diagnosed by adopting a maximum entropy multi-threshold segmentation method so as to obtain a human body part image contained in the kidney image to be diagnosed, and determining an optimal combined threshold of the maximum entropy multi-threshold segmentation method by adopting a particle swarm algorithm, so that the accuracy of kidney image segmentation is improved, and a foundation is laid for the subsequent tumor diagnosis;
(3) the particle swarm optimization is set, the iterative updating mode of the particle swarm is adjusted by adopting the cognitive learning factor and the social learning factor which change in a self-adaptive manner, so that the global exploration and the local exploitation of the particle swarm optimization are balanced, the optimizing precision and the convergence speed of the particle swarm optimization are improved, and the optimal combination threshold determined by adopting the particle swarm optimization has higher accuracy.
Drawings
The invention is further described with the aid of the accompanying drawings, in which, however, the embodiments do not constitute any limitation to the invention, and for a person skilled in the art, without inventive effort, further drawings may be derived from the following figures.
FIG. 1 is a schematic view of the present invention.
Detailed Description
The kidney disease automatic detection method and system based on machine learning comprise a kidney image collection module, a kidney image management module, a kidney influence intelligent diagnosis module and a kidney disease classification module;
kidney image collection module: for collecting images of various types of kidney diseases of patients, the sources of the kidney images include InterVar (site pathogenicity assessment), GeneRereviews (disease database), and images of the kidneys of patients in hospital.
Kidney image management module: the kidney image pre-storing unit is used for pre-storing various types of kidney images collected by the kidney image collecting module, screening the kidney images with kidney disease diagnosis labels through the kidney image updating unit, cleaning excessive noise, default or irrelevant images, and then updating a kidney image training set;
the intelligent kidney image diagnosis module comprises a kidney image processing unit and a kidney image database, wherein the kidney image processing unit extracts the disease characteristics of a kidney image by analyzing the kidney image with a kidney disease diagnosis label, a machine learning model is established by taking the kidney image updated by a kidney image updating unit as a training set, the kidney image to be judged is automatically detected for the disease, the trained kidney image with the kidney disease diagnosis label is stored in the kidney image database, and a distinguishable kidney image automatic disease detection model is generated so as to be convenient for next time without retraining and be directly used;
the kidney disease classification module further classifies the kidney diseases judged by the kidney image intelligent diagnosis module so as to ensure that a patient using the system can obtain a diagnosis and treatment scheme through self-checking kidney images.
Specifically, the kidney image collection module collects images from the sources of interavar (site pathogenicity assessment), GeneReviews (disease database) and the kidney images of the patients in the hospital, and collects the images to the sql database for online management.
Specifically, the kidney image updating unit screens kidney images with kidney disease diagnosis labels, wherein the labels comprise calcific foci, density of renal calyx and renal pelvis, renal medulla, cyst and cancer embolus.
Preferably, the medical image processing unit performs image segmentation on the medical image to be diagnosed by using a maximum entropy multi-threshold segmentation method, and determines an optimal combination threshold of the maximum entropy multi-threshold segmentation method used in the medical image processing unit by using a particle swarm algorithm.
Preferably, the fitness function of the particle swarm is set to be the maximum entropy, and the larger the fitness function of the particles in the particle swarm is, the better the solution obtained by particle optimization is.
Specifically, in the embodiment, the medical image to be diagnosed is subjected to image segmentation by using the maximum entropy multi-threshold segmentation method, so that a human body part image contained in the medical image to be diagnosed is obtained, and the optimal combined threshold of the maximum entropy multi-threshold segmentation method is determined by using the particle swarm algorithm, so that the accuracy of medical image segmentation is improved, and a foundation is laid for the subsequent tumor diagnosis.
Preferably, in the particle swarm algorithm, the particle swarm iteratively updates the solution of the particle in the following manner:
Vi(t+1)＝ωVi(t)+ci1(t)r1(Pbesti(t)-Xi(t))+ci2(t)r2(Gbest(t)-Xi(t))
Xi(t+1)＝Xi(t)+Vi(t+1)
in the formula, Xi(t +1) and Vi(t +1) denotes the solution and velocity, X, of particle i at the (t +1) th iteration, respectivelyi(t) and Vi(t) respectively represents the solution and velocity of particle i at the t-th iteration, ω represents an inertial weight factor, Pbesti(t) represents the historical optimal solution of particle i at the t-th iteration, Gbest (t) represents the global optimal solution of the particle swarm at the t-th iteration, r1And r2Representing a randomly generated random number between 0 and 1, ci1(t) cognitive learning factor for particle i at the t-th iteration, ci2(t) represents the social learning factor of particle i at the t-th iteration.
Preferably, the particle swarm is set to adjust the cognitive learning factor c of the particle i at the t-th iteration in the following wayi1(t) and social learning factor ci2The value of (t):
let omegai(t) represents the local neighborhood of particle i at the t-th iteration, and Ωi(t) is to solve Xi(t) A circular area with r (t) as a center and r (t) as a radius, wherein R (t) is a local neighborhood radius of the particle swarm in the t iteration, and the value of R (t) is set as follows:
Figure BDA0003641056930000101
wherein R isi(t) represents the local neighborhood radius of particle i at the t-th iteration, and
Figure BDA0003641056930000102
Figure BDA0003641056930000103
denotes the dissociation X of the particle population at the t-th iterationi(t) a solution for the first nearest particle, c represents a given positive integer, and c < N, where c may take a value of 10 and N represents the number of particles in the population; omegai,best(t) represents the historical optimal neighborhood of particle i at the t-th iteration, and Ωi,best(t) is to solve Pbesti(t) a circular area with r (t) as the center and r (t) as the radius, r (t) is the historical optimal neighborhood radius of the particle swarm in the t iteration, and
Figure BDA0003641056930000104
wherein r isi(t) represents the historical optimal neighborhood radius of particle i at the t-th iteration, an
Figure BDA0003641056930000105
In the formula,
Figure BDA0003641056930000106
denotes the dissociation of Pbest in the population at the t-th iterationi(t) a solution for the particle at a-th;
defining a function qi(t) for detecting the optimum symmetry of the particle i in the tth iteration, q is setiThe formula for (t) is:
Figure BDA0003641056930000107
in the formula, qi(t) represents the optimized uniformity of particle i at the t-th iteration, Ji(t) denotes the local optimum symmetry, S, of particle i at the t-th iterationi(t) represents the historical goodness of the particle i at the tth iteration, Ji(t) and SiThe values of (t) are:
Figure BDA0003641056930000108
Figure BDA0003641056930000109
in the formula, Xj(t) represents the solution of particle j at the t-th iteration, Xb(t) represents the solution of particle b at the t-th iteration, and Xj(t)≠Xb(t)，Xk(t) represents the solution of particle k at the t-th iteration, Xo(t) represents the solution of particle o at the t-th iteration, and Xk(t)≠Xo(t)，λi,j(t) is a determination function for the optimization between particle i and particle j at the t-th iteration, let fi(t) and fj(t) the fitness function values of particle i and particle j at the t-th iteration, respectively, are then
Figure BDA0003641056930000111
λi,b(t) is a determination function for the optimization between particle i and particle b at the t-th iteration, let fb(t) represents the fitness function value of particle b at the t-th iteration, then
Figure BDA0003641056930000112
mi,best(t) represents the historical optimum neighborhood Ωi,bestThe number of particles in (t);
when q isi(t)＜Hq1(t) or qi(t)＞Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure BDA0003641056930000113
Figure BDA0003641056930000114
when H is presentq1(t)≤qi(t)≤Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure BDA0003641056930000115
Figure BDA0003641056930000116
in the above formula, α and β are two constants for controlling ci1(t) and ci2(t) wherein α may take the value of 1, β may take the value of 0.5, Hq1(t) is the lower bound of the global optimization symmetry of the particle swarm at the t-th iteration, and
Figure BDA0003641056930000117
Hq2(t) is the upper limit value of the global optimizing uniformity of the particle swarm in the t iteration, and
Figure BDA0003641056930000118
wherein,
Figure BDA0003641056930000119
means representing the optimum symmetry of the particles in the population at the t-th iteration, an
Figure BDA00036410569300001110
TmaxGiven a maximum number of iterations.
Specifically, the kidney diseases judged by the intelligent kidney image diagnosis module are further classified by utilizing a back propagation algorithm so as to ensure that a patient using the system can obtain a diagnosis and treatment scheme through a self-checking kidney image, and the method comprises the following specific steps:
firstly, an M-layer neural network is built, and for the output of the (M +1) -th layer network, the following steps are provided:
am+1＝fm+1(Wm+1am+bm+1),m＝0,1,…,M-1
wherein M is the number of network layers, a is the output result of the neural network, namely the type of the kidney disease, and a exists0P and aM，am+1Is the output result of the (m +1) th layer neural network, amIs the output result of the mth layer neural network, fm+1Is the excitation function of the (m +1) th layer neural network, Wm+1Is the weight of the (m +1) th layer neural network, bm+1The bias of the (m +1) th layer neural network, assuming that there is a corresponding kidney disease type output t for each type of input p of the kidney disease characteristics, the following correspondence can be established:
{p1,t1},{p2,t2},…,{pq,tq},…,{pQ,tQ}
wherein p is1Is the input of the 1 st kidney image feature, t1Is the expected output, p, for the 1 st class2Is the input of the 2 nd kidney image feature, t2Is the expected output, p, for the 2 nd classqIs the input of the qth renal image feature, tqIs the expected output, p, for the qth classQIs the input of the Q-th kidney image feature, tQIs the desired output for the qth class, the algorithm will adjust the network parameters to minimize the mean squared error: f (x) E [ (t-a)2]＝E[(t-a)T(t-a)]Let E-t-a denote the error of the desired output from the predicted output, where F denotes the mean square error, x denotes the arguments of F, which in this case include t and a, E denotes the mathematical expectation, (t-a)TTranspose of the representation error by
Figure BDA0003641056930000121
Instead of f (x), the mean square error is calculated approximately:
Figure BDA0003641056930000122
wherein k is the kth iteration, and the steepest descent method of the approximate mean square error is as follows:
Figure BDA0003641056930000123
Figure BDA0003641056930000124
wherein a is a learning speed of the learning,
Figure BDA0003641056930000125
represents the weight of the ith element to the jth element at the (k +1) th iteration of the mth layer neural network,
Figure BDA0003641056930000126
and (3) representing the weight of the ith element to the jth element under the kth iteration of the mth layer neural network, and defining a chain rule as follows:
Figure BDA0003641056930000127
if f (n) is e ^ n and n is 2w, then f (n) (w) is e ^ {2w }, there are
Figure BDA0003641056930000128
Figure BDA0003641056930000129
Then use the chain rule
Figure BDA00036410569300001210
And
Figure BDA00036410569300001211
can be written as:
Figure BDA00036410569300001212
for the net input n of the mth layer, there are:
Figure BDA00036410569300001213
wherein s represents sensitivity, sm-1To representThe sensitivity of the (m-1) th layer neural network,
Figure BDA0003641056930000131
representing the net input of the ith element in an m-layer neural network,
Figure BDA0003641056930000132
represents the output of the jth element in an m-1 layer neural network,
Figure BDA0003641056930000133
represents the bias of the ith element in an m-layer neural network, thus:
Figure BDA0003641056930000134
if defined:
Figure BDA0003641056930000135
then
Figure BDA0003641056930000136
And
Figure BDA0003641056930000137
can be simplified as follows:
Figure BDA0003641056930000138
expressed as:
Figure BDA0003641056930000139
Figure BDA00036410569300001310
wherein,
Figure BDA00036410569300001311
representing the bias of the ith element in the (k +1) th iteration of the m-layer neural network, the matrix form can be expressed as:
Wm(k+1)＝wm(k)-αsm(am-1)T
bm(k+1)＝bm(k)-αsm
wherein, Wm(k +1) represents wmMatrix form of (k +1), i.e. set of weights after k +1 th iteration of the m-layer neural network, where smComprises the following steps:
Figure BDA00036410569300001312
wherein,
Figure BDA00036410569300001313
the net input to the mth layer representing the approximation error is partial,
Figure BDA00036410569300001314
the 1 st net input representing the approximation error to the mth layer is partial-derivative,
Figure BDA00036410569300001315
the 2 nd net input representing the approximation error to the mth layer is partial,
Figure BDA00036410569300001316
s-th layer representing approximation error pairmThe net inputs are subjected to partial derivation, and the Jacobian matrix is recorded as:
Figure BDA00036410569300001317
wherein,
Figure BDA0003641056930000141
representing the partial derivative of the net input to the layer m +1 neural network to the net input to the layer m neural network,
Figure BDA0003641056930000142
representing the 1 st Net input of the m +1 st layer neural network to the 1 st Net input of the m layer neural networkThe partial derivatives of (a) are,
Figure BDA0003641056930000143
representing the partial derivative of the 1 st net input of the (m +1) th layer neural network to the 2 nd net input of the m' th layer neural network,
Figure BDA0003641056930000144
represents the s of the 1 st net input of the m +1 st layer neural network to the m layer neural networkmThe partial derivative of the individual net inputs,
Figure BDA0003641056930000145
represents the partial derivative of the 2 nd net input of the m +1 th layer neural network to the 1 st net input of the m layer neural network,
Figure BDA0003641056930000146
represents the partial derivative of the 2 nd net input of the (m +1) th layer neural network to the 2 nd net input of the m layer neural network,
Figure BDA0003641056930000147
represents the s < nd > net input of the (m +1) th layer neural network to the m < th > neural networkmThe partial derivative of the individual net inputs,
Figure BDA0003641056930000148
s represents the (m +1) th layer neural networkm+1The partial derivative of this net input to the net input of the mth layer neural network,
Figure BDA0003641056930000149
s representing the (m +1) th neural networkm+1Partial derivatives of the 1 st net input to the mth layer neural network,
Figure BDA00036410569300001410
s representing the (m +1) th neural networkm+1S of net input to mth layer neural networkmThe partial derivative of the individual net inputs,
Figure BDA00036410569300001411
s represents the (m +1) th layer neural networkm+1S of net input to mth layer neural networkmPartial derivatives of the individual net inputs, consider the i, j elements of the matrix:
Figure BDA00036410569300001412
Figure BDA00036410569300001413
wherein:
Figure BDA00036410569300001414
then the jacobian matrix can be written as:
Figure BDA00036410569300001415
wherein:
Figure BDA00036410569300001416
Fm(nm) Represents the net input of the mth layer neural network as nmThe objective function of the time-of-day,
Figure BDA00036410569300001417
represents a net input to the mth layer neural network of
Figure BDA00036410569300001418
The objective function of the time of day,
Figure BDA00036410569300001419
represents a net input to the mth layer neural network of
Figure BDA00036410569300001420
The objective function of the time of day,
Figure BDA00036410569300001421
represents a net input to the mth layer neural network of
Figure BDA00036410569300001422
The objective function of time is then in the form of a matrixThe chain rule writes a recurrence relation of sensitivity:
Figure BDA00036410569300001423
Figure BDA00036410569300001424
specifically, the optimal training weight is obtained by using a back propagation algorithm and a chain method, and the disease information of the patient is input into a trained machine learning model, so that whether the patient has calcified foci, density of renal calyx and renal pelvis, renal medulla, cyst and cancer embolus diseases or not can be automatically detected.
Specifically, to the machine learning model who establishes, realize the kidney image automated inspection disease to needs are differentiated, will train good kidney image that has kidney disease diagnosis label and keep to kidney image database, generate the kidney image automated inspection disease's that can differentiate model and need not retraining once more in order to make things convenient for next time, directly use, kidney disease classification module is further categorised to the kidney disease that kidney image intelligent diagnosis module differentiateed to the patient of guaranteeing to use this system can obtain diagnosis and treatment scheme through self-checking kidney image.
Specifically, when the particle swarm algorithm is applied to determine the optimal combined threshold of the maximum entropy multi-threshold segmentation method, the diversity of the particle swarm is poor easily, so that the phenomenon of early maturity occurs, or the convergence speed of the particle swarm is slowed down while the diversity of the particle swarm is enhanced, so that the problem of resource waste occurs. That is, in the particle swarm optimization, increasing the diversity of the particles and increasing the convergence rate of the particles are contradictory, and the balance between the two is difficult. Considering that in the particle swarm algorithm, the cognitive learning factor maintains a larger value, which is helpful for searching particles in a large range, so as to improve the diversity of the particle swarm, but results in a slower convergence rate of the algorithm, while the social learning factor maintains a larger value, which is helpful for learning social experiences, so that the search result quickly converges to the vicinity of the optimal solution, so as to improve the convergence rate of the algorithm, but the search space capacity is weaker, therefore, the embodiment balances the global exploration and the local exploitation of the particle swarm algorithm by adopting the cognitive learning factor and the social learning factor value which are adaptively changed, thereby playing roles of improving the diversity of the particle swarm and improving the convergence rate. In the iterative updating formula of the particles, the cognitive part reflects the self-learning ability of the particles in exploration and reflects the following trend of the particles to the best characteristics of the particles, and the social part reflects the ability of the particles to learn from the outside and reflects the group behaviors of cooperative cooperation and knowledge sharing among the particles. And the cognitive learning factor and the social learning factor respectively determine the contribution degree of the cognitive part and the social part in the particle iterative updating process. Considering that when the particles in the particle swarm optimize the current solution space more uniformly, the diversity of the particle swarm is better, and the globally optimal solution is found with a higher probability, therefore, the embodiment defines the optimization symmetry to be used for measuring the uniformity of the particle in the solution space where the particle is located, when the uniformity of the particle in the solution space where the particle is located is measured, not only the uniformity of the solution space where the particle is located is measured by the local optimization symmetry in the optimization symmetry, but also the uniformity of the solution space where the particle is located is measured by the historical optimization symmetry in the optimization symmetry, so that when the value of the cognitive learning factor of the particle is increased according to the optimization symmetry of the particle, the cognitive part of the particle can better increase the diversity of the particle. In the particle swarm, when the individual optimizing symmetry of the particle is smaller or larger than the global optimizing symmetry, it indicates that the particle is not sufficiently learning the solution space it is in the current exploration compared to the global, thus, increasing the cognitive learning factor of the particle, decreasing the social learning factor of the particle, namely, the contribution degree of the cognitive part in the iterative updating process is enhanced, so that the diversity of the algorithm is improved, when the optimizing uniformity of the particle individual is moderate compared with the global optimizing uniformity, it indicates that the particle has learned the solution space in which it is located more fully than globally in the current exploration, thus, decreasing the cognitive learning factor of the particle, increasing the social learning factor of the particle, namely, the contribution degree of the social part of the algorithm in the iterative updating process is enhanced, so that the convergence speed of the algorithm is improved. The convergence rate of the particle swarm algorithm can be improved under the condition that the diversity of the particle swarm in the current solution space is better ensured through the cognitive learning factor and the social learning factor regulated by the optimizing uniformity, and therefore the optimizing precision of the particle swarm is improved.
The beneficial effects of the invention are as follows: the artificial intelligence technology is applied to kidney image analysis, tumor diagnosis models corresponding to human body parts based on kidney images are established, tumor diagnosis can be effectively carried out on the human body parts, and the functions of automatic analysis of the kidney images and medical tumor diagnosis assisting doctors are realized; the method comprises the steps of performing image segmentation on a kidney image to be diagnosed by adopting a maximum entropy multi-threshold segmentation method so as to obtain a human body part image contained in the kidney image to be diagnosed, and determining an optimal combined threshold of the maximum entropy multi-threshold segmentation method by adopting a particle swarm algorithm, so that the accuracy of kidney image segmentation is improved, and a foundation is laid for the subsequent tumor diagnosis; the particle swarm optimization is set, the iterative updating mode of the particle swarm is adjusted by adopting the cognitive learning factor and the social learning factor which change in a self-adaptive manner, so that the global exploration and the local exploitation of the particle swarm optimization are balanced, the optimizing precision and the convergence speed of the particle swarm optimization are improved, and the optimal combination threshold determined by adopting the particle swarm optimization has higher accuracy.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the protection scope of the present invention, although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims (10)

1. The kidney disease automatic detection method and system based on machine learning comprise a kidney image collection module, a kidney image management module, a kidney influence intelligent diagnosis module and a kidney disease classification module;
kidney image collection module: the system is used for collecting kidney disease images of various types of patients, wherein the sources of the kidney images comprise InterVar (site pathogenicity judgment), GeneReviews (disease database) and kidney images of patients in a hospital;
kidney image management module: the kidney image pre-storing unit is used for pre-storing various types of kidney images collected by the kidney image collecting module, screening the kidney images with kidney disease diagnosis labels through the kidney image updating unit, and updating a kidney image training set after the kidney images with excessive cleaning noise and default or irrelevant images are screened;
the intelligent kidney image diagnosis module comprises a kidney image processing unit and a kidney image database, wherein the kidney image processing unit extracts the disease characteristics of a kidney image by analyzing the kidney image with a kidney disease diagnosis label, a machine learning model is established by taking the kidney image updated by a kidney image updating unit as a training set, the kidney image to be judged is automatically detected for the disease, the trained kidney image with the kidney disease diagnosis label is stored in the kidney image database, and a distinguishable kidney image automatic disease detection model is generated so as to be convenient for next time without retraining and be directly used;
the kidney disease classification module further classifies the kidney diseases judged by the kidney image intelligent diagnosis module so as to ensure that a patient using the system can obtain a diagnosis and treatment scheme through a self-checking kidney image.
2. The method and system for automatic kidney disease detection based on machine learning of claim 1, wherein the kidney image collection module collects from sources of InterVar (site pathogenicity assessment), GeneReviews (disease database) and kidney images of patients in hospital and collects them to be collected to the sql database for online management.
3. The method and system for automatic detection of kidney diseases based on machine learning as claimed in claim 1, wherein the kidney image updating unit screens the kidney images with diagnosis labels of kidney diseases, the labels including calcific foci, density of renal calyx, renal medulla, cyst and cancer embolus.
4. The method and system for automatic detection of kidney diseases based on machine learning of claim 1, wherein a kidney image processing unit performs image segmentation on the kidney image to be diagnosed by using a maximum entropy multi-threshold segmentation method, and determines an optimal combination threshold of the maximum entropy multi-threshold segmentation method used in the kidney image processing unit by using a particle swarm optimization.
5. The method and system for machine learning based automated detection of kidney disease of claim 4, wherein in the particle swarm algorithm, the particle swarm iteratively updates the solution of the particle in the following way:
Vi(t+1)＝ωVi(t)+ci1(t)r1(Pbesti(t)-Xi(t))+ci2(t)r2(Gbest(t)-Xi(t))
Xi(t+1)＝Xi(t)+Vi(t+1)
in the formula, Xi(t +1) and Vi(t +1) denotes the solution and velocity of particle i at the (t +1) th iteration, X, respectivelyi(t) and Vi(t) respectively represents the solution and velocity of the particle i at the t-th iteration, ω represents an inertial weight factor, Pbesti(t) represents the historical optimal solution of particle i at the t-th iteration, Gbest (t) represents the global optimal solution of the particle swarm at the t-th iteration, r1And r2Representing a randomly generated random number between 0 and 1, ci1(t) cognitive learning factor for particle i at the t-th iteration, ci2(t) represents the social learning factor of particle i at the tth iteration.
6. The method and system for automatic detection of kidney disease based on machine learning of claim 5, wherein the particle group is configured to adjust the cognitive learning factor c of the particle i at the t-th iteration in the following manneri1(t) and social learning factor ci2Value of (t):
let omegai(t) represents particlesLocal neighborhood of sub i at the t-th iteration, and Ωi(t) is to solve Xi(t) a circular region with R (t) as the center and R (t) as the radius, R (t) is the local neighborhood radius of the particle swarm in the t iteration, and the value of R (t) is set as follows:
Figure FDA0003641056920000021
wherein R isi(t) represents the local neighborhood radius of particle i at the t-th iteration, and
Figure FDA0003641056920000022
Figure FDA0003641056920000023
denotes the distance dissociation X in the particle population at the t-th iterationi(t) a solution for the first nearest particle, c representing a given positive integer, and c < N, N representing the number of particles in the population; omegai，best(t) represents the historical optimum neighborhood of particle i at the t-th iteration, and Ωi，best(t) is to solve Pbesti(t) a circular area with r (t) as the center and r (t) as the radius, r (t) is the historical optimal neighborhood radius of the particle swarm in the t iteration, and
Figure FDA0003641056920000024
wherein r isi(t) represents the historical optimal neighborhood radius of particle i at the t-th iteration, an
Figure FDA0003641056920000025
In the formula,
Figure FDA0003641056920000026
denotes the dissociation of Pbest in the population at the t-th iterationi(t) a solution for the particle at a-th;
defining a function qi(t) for detecting the optimum symmetry of the particle i in the tth iteration, q is setiThe formula for (t) is:
Figure FDA0003641056920000027
in the formula, qi(t) shows the optimum uniformity, J, of particle i at the tth iterationi(t) denotes the local optimum symmetry, S, of particle i at the t-th iterationi(t) represents the historical goodness of the particle i at the tth iteration, Ji(t) and SiThe values of (t) are:
Figure FDA0003641056920000031
Figure FDA0003641056920000032
in the formula, Xj(t) represents the solution of particle j at the t-th iteration, Xb(t) represents the solution of particle b at the tth iteration, and Xj(t)≠Xb(t)，Xk(t) represents the solution of particle k at the t-th iteration, Xo(t) represents the solution of particle o at the t-th iteration, and Xk(t)≠Xo(t)，λi，j(t) is the function of the optimization decision between particle i and particle j at the t-th iteration, let fi(t) and fj(t) the fitness function values of particle i and particle j at the t-th iteration, respectively, are then
Figure FDA0003641056920000033
λi，b(t) is a determination function for the optimization between particle i and particle b at the t-th iteration, let fb(t) represents the fitness function value of particle b at the t-th iteration, then
Figure FDA0003641056920000034
mi，best(t) represents the historical optimum neighborhood Ωi，bestThe number of particles in (t);
when q isi(t)＜Hq1(t) or qi(t)＞Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure FDA0003641056920000035
Figure FDA0003641056920000036
when H is presentq1(t)≤qi(t)≤Hq2At (t), then ci1(t) and ci2The values of (t) were adjusted to:
Figure FDA0003641056920000037
Figure FDA0003641056920000038
in the above formula, α and β are two constants for controlling ci1(t) and ci2(t) range, Hq1(t) is the lower bound of the global optimum symmetry of the particle swarm at the t-th iteration, and
Figure FDA0003641056920000041
Hq2(t) is the upper limit value of the global optimizing uniformity of the particle swarm in the t iteration, and
Figure FDA0003641056920000042
wherein,
Figure FDA0003641056920000043
means representing the optimum symmetry of the particles in the population at the t-th iteration, an
Figure FDA0003641056920000044
TmaxGiven a maximum number of iterations.
7. The machine learning-based method and system for automatically detecting kidney diseases according to claim 1, wherein the kidney diseases identified by the intelligent kidney image diagnosis module are further classified by using a back propagation algorithm to ensure that a patient using the system can obtain a diagnosis and treatment plan through self-examination of the kidney images, and the method comprises the following specific steps:
firstly, an M-layer neural network is built, and for the output of the (M +1) -th layer network, the following steps are provided:
am+1＝fm+1(Wm+1am+bm+1)，m＝0，1，...，M-1
wherein M is the number of network layers, a is the output result of the neural network, namely the type of the kidney disease, and a exists0P and aM，am+1Is the output result of the (m +1) th layer neural network, amIs the output result of the mth layer neural network, fm+1Is the excitation function of the neural network of layer m +1, Wm+1Is the weight of the (m +1) th neural network, bm+1Is the bias of the (m +1) th layer neural network, assuming that for each type of input p of the kidney disease characteristics, there is a corresponding kidney disease type output t, the following correspondence can be established: { p1，t1}，{p2，t2}，...，{pq，tq}，...，{pQ，tQIn which p is1Is the input of the 1 st kidney image feature, t1Is the expected output, p, for the 1 st class2Is the input of the 2 nd kidney image feature, t2Is the expected output, p, for the 2 nd classqIs the input of the qth renal image feature, tqIs the expected output, p, corresponding to the qth classQIs the input of the Q-th renal image feature, tQIs the desired output for the qth class, the algorithm will adjust the network parameters to minimize the mean squared error: f (x) E [ (t-a)2]＝E[(t-a)T(t-a)]Let e-t-a denote the error between the desired output and the predicted output, where F denotes the mean square error and x denotes the mean square errorF, when the independent variables include t and a, E represents the mathematical expectation, (t-a)TTranspose of the representation error by
Figure FDA0003641056920000045
Instead of f (x), the mean square error is calculated approximately:
Figure FDA0003641056920000046
Figure FDA0003641056920000047
wherein k is the kth iteration, and the steepest descent method of the approximate mean square error is as follows:
Figure FDA0003641056920000048
wherein a is a learning speed of the learning,
Figure FDA0003641056920000049
represents the weight of the ith element to the jth element at the (k +1) th iteration of the mth layer neural network,
Figure FDA0003641056920000051
representing the weight of the ith element to the jth element at the kth iteration of the mth layer neural network.
8. The method and system for machine learning based automated renal disease detection according to claim 7, wherein a chain rule is defined as:
Figure FDA0003641056920000052
if f (n) is e ^ n and n is 2w, then f (n) (w) is e ^ 2w, have
Figure FDA0003641056920000053
Figure FDA0003641056920000054
Then the chain rule is utilized
Figure FDA0003641056920000055
And
Figure FDA0003641056920000056
can be written as:
Figure FDA0003641056920000057
for the net input n of the mth layer, there are:
Figure FDA0003641056920000058
wherein s represents sensitivity, sm-1Representing the sensitivity of the (m-1) th layer neural network,
Figure FDA0003641056920000059
representing the net input of the ith element in an m-layer neural network,
Figure FDA00036410569200000510
represents the output of the jth element in an m-1 layer neural network,
Figure FDA00036410569200000511
represents the bias of the ith element in an m-layer neural network, thus:
Figure FDA00036410569200000512
if defined:
Figure FDA00036410569200000513
then the
Figure FDA00036410569200000514
And
Figure FDA00036410569200000515
can be simplified as follows:
Figure FDA00036410569200000516
expressed as:
Figure FDA00036410569200000517
Figure FDA00036410569200000518
wherein,
Figure FDA00036410569200000519
representing the bias of the ith element in the (k +1) th iteration of the m-layer neural network, the matrix form can be expressed as:
Wm(k+1)＝wm(k)-αsm(am-1)T
bm(k+1)＝bm(k)-αsm
wherein, Wm(k +1) represents wm(k +1) matrix form, i.e. set of weights after k +1 iterations of the m-layer neural network, where smComprises the following steps:
Figure FDA00036410569200000520
wherein,
Figure FDA0003641056920000061
the net input to the mth layer representing the approximation error is partial,
Figure FDA0003641056920000062
the 1 st net input representing the approximation error to the mth layer is partial,
Figure FDA0003641056920000063
indicating approximation errorThe 2 nd net input to the mth layer is partial-derivative,
Figure FDA0003641056920000064
s of the m-th layer representing approximation errormThe net inputs are subjected to partial derivation, and the Jacobian matrix is recorded as:
Figure FDA0003641056920000065
wherein,
Figure FDA0003641056920000066
represents the partial derivative of the net input to the layer m +1 neural network to the net input to the layer m neural network,
Figure FDA0003641056920000067
represents the partial derivative of the 1 st net input of the (m +1) th layer neural network to the 1 st net input of the m' th layer neural network,
Figure FDA0003641056920000068
represents the partial derivative of the 1 st net input of the m +1 th layer neural network to the 2 nd net input of the m layer neural network,
Figure FDA0003641056920000069
represents the s-th net input of the (m +1) -th layer neural network to the (m-th layer) neural networkmThe partial derivative of the individual net inputs,
Figure FDA00036410569200000610
represents the partial derivative of the 2 nd net input of the m +1 th layer neural network to the 1 st net input of the m layer neural network,
Figure FDA00036410569200000611
representing the partial derivative of the net 2 input of the (m +1) th layer neural network to the net 2 input of the mth layer neural network,
Figure FDA00036410569200000612
represents the s < nd > net input of the (m +1) th layer neural network to the m < th > neural networkmThe partial derivative of the individual net inputs,
Figure FDA00036410569200000613
s represents the (m +1) th layer neural networkm+1The partial derivative of the net input to the net input of the mth layer neural network,
Figure FDA00036410569200000614
s represents the (m +1) th layer neural networkm+1Partial derivatives of the 1 st net input to the mth layer neural network,
Figure FDA00036410569200000615
s representing the (m +1) th neural networkm+1S of net input to mth layer neural networkmThe partial derivative of the individual net inputs,
Figure FDA00036410569200000616
s represents the (m +1) th layer neural networkm+1S of net input to mth layer neural networkmPartial derivatives of the individual net inputs, consider the i, j elements of the matrix:
Figure FDA00036410569200000617
Figure FDA00036410569200000618
wherein:
Figure FDA00036410569200000619
then the jacobian matrix can be written;
Figure FDA00036410569200000620
wherein;
Figure FDA0003641056920000071
Fm(nm) Represents the net input of the mth layer neural network as nmThe objective function of the time-of-day,
Figure FDA0003641056920000072
represents a net input to the mth layer neural network of
Figure FDA0003641056920000073
The objective function of the time-of-day,
Figure FDA0003641056920000074
represents a net input to the mth layer neural network of
Figure FDA0003641056920000079
The objective function of the time-of-day,
Figure FDA0003641056920000075
represents a net input to the mth layer neural network of
Figure FDA0003641056920000076
And (3) writing a sensitive recurrence relation through a chain rule of a matrix form by using a time objective function:
Figure FDA0003641056920000077
Figure FDA0003641056920000078
9. the method and system for automatic detection of kidney diseases based on machine learning of claim 8, wherein the optimal training weight is obtained by using back propagation algorithm and chain method, and the patient's disease information is inputted into the trained machine learning model, so as to automatically detect whether the patient has calcific foci, density of renal pelvis, renal medulla, cyst and cancer embolus.
10. The method and system for automatically detecting kidney diseases based on machine learning according to claim 1, wherein for the established machine learning model, the kidney images to be determined are automatically detected for diseases, the trained kidney images with the diagnosis labels for kidney diseases are stored in a kidney image database, a model for automatically detecting diseases with kidney images capable of being determined is generated for the convenience of direct use without retraining next time, and the kidney disease classification module further classifies the kidney diseases determined by the intelligent diagnosis module for kidney images, so as to ensure that a patient using the system can obtain a diagnosis and treatment plan through the self-diagnosis kidney images.
CN202210514808.6A 2022-05-12 2022-05-12 Method and system for automatically detecting kidney diseases based on machine learning Pending CN114783593A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210514808.6A CN114783593A (en) 2022-05-12 2022-05-12 Method and system for automatically detecting kidney diseases based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210514808.6A CN114783593A (en) 2022-05-12 2022-05-12 Method and system for automatically detecting kidney diseases based on machine learning

Publications (1)

Publication Number Publication Date
CN114783593A true CN114783593A (en) 2022-07-22

Family

ID=82436998

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210514808.6A Pending CN114783593A (en) 2022-05-12 2022-05-12 Method and system for automatically detecting kidney diseases based on machine learning

Country Status (1)

Country Link
CN (1) CN114783593A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI802486B (en) * 2022-08-15 2023-05-11 臺中榮民總醫院 Method of Interpreting Ultrasonic Images of Children's Kidneys Using Artificial Intelligence

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112530537A (en) * 2020-12-15 2021-03-19 罗鑫龙 Big health management platform based on algorithm, medical image and block chain
WO2021055596A1 (en) * 2019-09-17 2021-03-25 Quadrus Medical Technologies, Inc. System and method for personalized kidney evaluation, diagnosis and therapy recommendation
CN113870999A (en) * 2020-12-15 2021-12-31 广东德澳智慧医疗科技有限公司 Remote disease intelligent diagnosis system and auxiliary diagnosis method based on algorithm, medical image and block chain

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021055596A1 (en) * 2019-09-17 2021-03-25 Quadrus Medical Technologies, Inc. System and method for personalized kidney evaluation, diagnosis and therapy recommendation
CN112530537A (en) * 2020-12-15 2021-03-19 罗鑫龙 Big health management platform based on algorithm, medical image and block chain
CN113870999A (en) * 2020-12-15 2021-12-31 广东德澳智慧医疗科技有限公司 Remote disease intelligent diagnosis system and auxiliary diagnosis method based on algorithm, medical image and block chain

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI802486B (en) * 2022-08-15 2023-05-11 臺中榮民總醫院 Method of Interpreting Ultrasonic Images of Children's Kidneys Using Artificial Intelligence

Similar Documents

Publication Publication Date Title
Dybowski et al. Clinical applications of artificial neural networks
CN108351862A (en) The method and apparatus for determining development progress using artificial intelligence and user&#39;s input
JP2006518062A (en) Prediction algorithm training and testing database optimization system and method
Zhao et al. A modified artificial immune system based pattern recognition approach—An application to clinical diagnostics
CN113113130A (en) Tumor individualized diagnosis and treatment scheme recommendation method
Elhariri et al. Using hybrid filter-wrapper feature selection with multi-objective improved-salp optimization for crack severity recognition
CN115060497A (en) Bearing fault diagnosis method based on CEEMD energy entropy and optimized PNN
Chatterjee et al. Hybrid non-dominated sorting genetic algorithm: II-neural network approach
CN112652361A (en) High-risk myeloma screening method based on GBDT model and application thereof
CN113610118B (en) Glaucoma diagnosis method, device, equipment and method based on multitasking course learning
Mathew et al. A multimodal adaptive approach on soft set based diagnostic risk prediction system
Sindhiya et al. A survey on genetic algorithm based feature selection for disease diagnosis system
Phankokkruad Cost-sensitive extreme gradient boosting for imbalanced classification of breast cancer diagnosis
CN114783593A (en) Method and system for automatically detecting kidney diseases based on machine learning
Anand et al. HANN: a hybrid model for liver syndrome classification by feature assortment optimization
Prasanna et al. Heart Disease Prediction using Reinforcement Learning Technique
Baviskar et al. Efficient heart disease prediction using hybrid deep learning classification models
Seenivasagam et al. MYOCARDIAL INFARCTION DETECTION USING INTELLIGENT ALGORITHMS.
Dinesh et al. Reliable evaluation of neural network for multiclass classification of real-world data
Kasim Ensemble classification based optimized transfer learning feature method for early stage diagnosis of diabetic retinopathy
Rashid et al. A convolutional neural network based classification approach for breast Cancer detection
Krishnamoorthy et al. AO-HRCNN: archimedes optimization and hybrid region-based convolutional neural network for detection and classification of diabetic retinopathy
Behera et al. Optimizing Multi-Layer Perceptron using Variable Step Size Firefly Optimization Algorithm for Diabetes Data Classification.
Kassylkassova et al. Automated Pneumonia Diagnosis using a 2D Deep Convolutional Neural Network with Chest X-Ray Images
Pareek et al. Prediction of CKD using expert system fuzzy logic & AI

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
CB02 Change of applicant information
CB02 Change of applicant information

Address after: Dianchi Lingxiu Community, Liujia Sub District Office, Guandu District, Kunming City, Yunnan Province

Applicant after: Luo Xuemin

Address before: 730000 No. 222 Tianshui South Road, Gansu, Lanzhou

Applicant before: Luo Xuemin

TA01 Transfer of patent application right
TA01 Transfer of patent application right

Effective date of registration: 20240308

Address after: Room 501, Room 502, room 503, room 504, room 505, floor 5, building 10, Industrialization Center, No.19 Alishan Road, Songshanhu high tech Industrial Development Zone, Dongguan City, Guangdong Province 523000

Applicant after: Guangdong deao smart Medical Technology Co.,Ltd.

Country or region after: China

Address before: Dianchi Lingxiu Community, Liujia Sub District Office, Guandu District, Kunming City, Yunnan Province

Applicant before: Luo Xuemin

Country or region before: China

CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Jin Haijiao

Inventor after: Lin Qisheng

Inventor after: Li Xiaoyang

Inventor after: Luo Xuemin

Inventor after: Wang Hongping

Inventor before: Luo Xuemin

Inventor before: Wang Hongping