WO2022121063A1

WO2022121063A1 - Method for optimization of disease analysis and identification by multi-data correlation fusion of characterization, blood, and medical image data

Info

Publication number: WO2022121063A1
Application number: PCT/CN2021/000232
Authority: WO
Inventors: 谈斯聪; 于皓; 于梦非
Original assignee: 谈斯聪; 于皓; 于梦非
Priority date: 2020-12-11
Filing date: 2021-12-08
Publication date: 2022-06-16
Also published as: AU2021393938A1

Abstract

Method for optimization of disease analysis and identification by multi-data correlation fusion of characterization, blood, and medical image data using artificial intelligence theory, and smart recognition and smart analysis technology. Using an improved correlation method, the diagnostic gain is calculated by summing the probabilities of confirmed diagnosis for various types of features, and a genetic optimization method is improved. Confirmation conditions are satisfied, and diagnostic gain is optimized. The optimal combination of feature data for each minimized diagnostic gain is obtained. By utilizing artificial intelligence and optimization theory, problems such as the application of single-data diagnosis, low diagnostic efficiency, and inaccurate diagnosis are solved, and problems such as unnecessary medical data affecting medical diagnosis, resulting in medical diagnosis and analysis being inaccurate. Using multi-data fusion, correlation, decimation of blood data, image data, and characterization feature correlation rules, optimization theory is applied to decimate a combination of features for multi-data fusion, and the combination of features for multi-data fusion is calculated, increasing the efficiency of inference analysis.

Description

A disease analysis and identification optimization method based on multi-data association fusion of representation, blood and medical image data

Technical field:

The invention belongs to the technical field of artificial intelligence, and relates to various types of medical data analysis technology, image intelligent recognition technology, optimization method and other technologies and methods.

Background technique:

Currently used in the field of artificial intelligence; due to the variety of medical data, including CT, ultrasound, nuclear magnetic resonance and other image data, as well as blood data, organ characterization and other data. There are many types of data, and multi-data association and fusion have become the key to data analysis and intelligent diagnosis. At present, single-item data diagnosis is widely used, and the diagnosis efficiency is low and inaccurate. The fusion of various types of data is time-consuming and labor-intensive. Unnecessary medical data items will affect judgment and lead to problems such as inaccurate medical diagnosis and analysis. Using characterization, blood, image data multi-data association methods, multi-data fusion analysis, diagnosis, classification and identification of abnormal indicators. Calculate the sum of the identification probability of the characteristic item, the diagnostic gain improves the genetic optimization method, and according to the diagnostic gain of the characteristic item described in S12 of claim 1, the genetic optimization method is used to judge the characteristic optimal characteristic group of the disease, intelligently identify the disease, and realize Diagnostic data performance is optimized.

Invention content:

The purpose of the present invention is to provide an optimal method for disease analysis and identification of characterization, blood, and image data multi-data association fusion. The combination of artificial intelligence technology and various types of medical data correlation analysis technology and intelligent diagnosis improves the efficiency of data analysis and judgment and improves the rate of disease diagnosis.

Using representation, blood, and image data multi-data fusion association, the main organ is associated with related organs, and the association rules of various types of medical data between organs are analyzed. Using the sum of the identification probabilities of the feature items, an improved genetic method is used to optimize the identification probability of the feature items. Analyze various abnormal indicators, effectively identify medical images, intelligently manage medical data, realize the optimal combination of medical data diagnosis items and feature items, and improve the efficiency of analysis and diagnosis.

The technical scheme adopted in the present invention:

A characterization, blood, image data multi-data fusion disease analysis and identification optimization method

S0. Create a disease initial feature set of the main organ, and initialize the recognition probability;

S1. Create a list of associated organ objects of the main organ;

S2. Extract the feature items of the associated organs of the main organ;

S3. Add an abnormal feature item of an organ to the list of organ objects;

S4. Read abnormal blood indexes, screen and analyze, and add abnormal feature items to the feature candidate list;

S5. Characterize the association between organs and other organs, and perform data association analysis. Add abnormal feature items to the feature candidate list;

S6. Abnormal images of blood cells in other organs are read, and images of other organs and tissues are abnormal;

Add abnormal feature items and values to the feature candidate list;

S7, create a list of optimal feature groups, extract elements that meet the requirements and put them into the optimal feature group;

S8. Traverse all feature items in the data set, and randomly extract feature items as a division method,

The sum of the recognition probabilities of all feature items of each division method is taken as the total recognition probability of each division method;

S9. Calculate the information gain calculation method: diagnostic gain = total identification probability of each division mode - initialization identification probability;

S10. Constraints are: the diagnostic gain of the division method > the fixed identification index (fixed identification index=1);

S11, the objective function is to minimize the diagnostic gain;

S12. The optimal calculation method obtains the optimal diagnosis data set of the disease according to the optimization method.

Further, the identification probability sum calculation method of the feature item described in S8 is as follows:

For each disease, calculate the identification probability calculation method corresponding to each feature item and the disease:

Step1. To characterize each sample, extract its corresponding organ;

Step2. Extract its organ image data, blood cell image data and blood data in the sample;

Step3. If it is a blood abnormal feature item, increase the count value of the feature item. Increase the count value of the total feature item;

Step4. For each type of organ, for the blood cell image of each organ,

If it is a blood abnormal feature item, increase the count value of the feature item,

Increase the count value of the total feature item;

Step5. For each type of organ,

For each organ tissue image,

Increase the count value of the total feature item;

Step6. If the abnormal feature count value of the feature > the identification number of the corresponding feature item, set it as the hard identification probability,

If, the abnormal feature count value of the feature < the corresponding identification number, set it as the soft identification probability;

Step7. Return the recognition probability of the feature.

Further, the identification probability of the characteristic item described in S12 is optimized to improve the genetic method as follows:

Step1. Set each division method as a population;

Step2. Select the feature items of each division method as genes, and the total number of feature items as the number of genes;

Step3. Variable setting, when the feature item is selected in this division method, the variable value of the selected locus is 1;

The variable value of the unselected locus is 0;

Step4. Use to select the optimal population;

Step5. In the cross parent chromosomes A and B, the gene method of selecting two genes and exchanging loci is:

"11" in a chromosome is exchanged with "01", "10" and "00" in another chromosome, and the method of exchange is not limited to this;

Step6. Gene mutation. Select the locus with the smallest identification probability value and change 1 to 0;

Step7. The objective function is to minimize the diagnostic gain.

To sum up, the beneficial effects of the present invention are:

Aiming at the problems of low efficiency, time-consuming and labor-intensive operation of medical analysis and judgment, the invention solves the problems in the prior art, such as difficulty in data analysis and inaccurate data judgment, through multi-data fusion.

Through a single data indicator, it is difficult to effectively identify anomalies in a single type of data. A characterization, blood, image data multi-data correlation fusion disease analysis and diagnosis optimization method efficiently integrates multi-type data, multi-organ data, correlation analysis and comprehensive data, and realizes the optimal feature data set.

Description of drawings:

FIG. 1 is a schematic flowchart of the multi-data fusion disease analysis and diagnosis optimization method in the specification of the present application.

FIG. 2 is a schematic flow chart of the improved genetic method for optimizing the feature set in the specification of the present application.

Example 1:

As shown in FIG. 1 and FIG. 2 , a characterization, blood, image data multi-data association fusion disease analysis and diagnosis optimization method steps method and its implementation method are taken as an example, but the implementation method is not limited to this. The step method of the embodiment is as follows:

Input the disease characteristics of the main organ, and initialize the recognition probability. Enter a representation to create a list of associated organ objects for the primary organ.

Extract the associated organs of the main organ and the feature items corresponding to the associated organs. Add Organ Abnormal Feature to the list of Organ Objects.

Read the abnormal blood index, and add the abnormal feature item to the feature candidate list. Establish associations between representative organs and other organs, and data association analysis. Add anomalous feature items to the feature candidate list. The images of blood cells in other organs are abnormal, and the images of other organs and tissues are abnormal. Add anomalous feature items and their values to the feature candidate list. Create a list of optimal feature groups. Extract the elements that meet the requirements and put them into the optimal feature group.

For each disease, the identification probability corresponding to each feature item and the disease is calculated, the representation of the sample is calculated and extracted, and the corresponding organ is extracted. Its organ image data, blood cell image data and blood data in the sample. If there is a blood abnormal feature item, increase the count value of the feature item. Increment the count value of the total feature item. For each type of organ, for the blood cell image of each organ, if it is a blood cell abnormal feature item, increase the count value of the feature item. Increment the count value of the total feature item. For each type of organ, for each organ tissue image, if it is an abnormal feature item of the organ tissue, increase the count value of the feature item. Increment the count value of the total feature item. If the abnormal feature count value of the feature > the identification number of the corresponding feature item, it is set as the hard identification probability. If the abnormal feature count value of the feature < the corresponding number of recognitions, it is set as the soft recognition probability. Returns the recognition probability for this feature.

Traverse all feature items in the data set, randomly extract feature items as a division method, and calculate the sum of the recognition probabilities of all feature items in each division method as the total recognition probability of each division method. Calculate the information gain of each division method, and apply the total recognition probability of each division method - the initial recognition probability. Constraints are determined when the diagnostic gain of the division method>fixed identification index (fixed identification index=1). If the conditions are met, if the conditions are not met, return to the upper loop. Enter the calculation iteration and find the optimal group of features that minimizes the gain.

Set each division as a chromosome. The feature items of each division method are selected as genes, and the total number of feature items is used as the number of genes. Set the variable. When the feature item is selected in this division method, the variable value of the selected locus is 1. The variable value of the unselected locus is 0. Use to select the optimal population. Cross parent chromosomes A, B. The gene method of selecting two genes and exchanging loci is as follows: "11" in a chromosome is exchanged with "01", "10" and "00" in another chromosome. The method of exchange is not limited to this. Gene mutation. Select the locus with the smallest identification probability value and change 1 to 0. Enter the iteration and find the objective function to minimize the diagnostic gain. The parameters include; the population size ranges from 128 to 256. The crossover rate was between 0.01-0.75, the mutation rate was between 0.01-0.05, and the number of iterations was between 2000-5000.

Claims

A disease analysis and identification optimization method based on multi-data association and fusion of characterization, blood and medical image data, which is characterized by merging various types of data and multi-feature data, including blood data, each organ characterization data, blood cell images, and organ tissue medical treatment. image data. Multi-data and multi-organ correlation, analysis, and calculation of optimal disease intelligence recognition gain as the optimal method for the objective function. Proceed as follows:

S1. Create a disease initial feature set of the main organ, and initialize the disease intelligent recognition probability;

S2. Create a list of associated organ objects of the main organ;

S3. Extract the feature items of the associated organs of the main organ;

S4. Add an abnormal feature item of an organ to the list of organ objects;

S5. Read abnormal blood indexes, screen and analyze, and add abnormal feature items to the feature candidate list;

S6. Characterize the association of organs and other organs, perform data association analysis, and add abnormal feature items to the feature candidate list;

S7. The images of blood cells in other organs are abnormal, and the images of other organs and tissues are abnormal;

Add abnormal feature items and values to the feature candidate list;

S8. Create an optimal feature group list. Extract the elements that meet the requirements and put them into the optimal feature group;

S9. Traverse all the feature items in the data set, and randomly extract the feature items as the division method,

The sum of the probability of intelligent identification of diseases of all feature items of each division method is taken as the total probability of intelligent identification of diseases of each division mode,

Set the values of hard disease intelligent recognition probability and soft disease intelligent recognition probability as parameters for calculating the total probability of disease intelligent recognition;

S10. Calculate the intelligent disease identification gain, and the calculation method is as follows: the intelligent disease identification gain = the total probability of intelligent identification of diseases of each division method - the probability of intelligent identification of diseases initialized;

S11. Constraints are: the disease intelligent identification gain of the division method > the fixed disease intelligent identification index (fixed disease intelligent identification index=1);

S12, the objective function is to minimize the gain of intelligent disease identification;

S13, the optimal calculation method obtains the optimal disease intelligent identification data set of the disease according to the optimal method;
A disease analysis and identification optimization method based on multi-data association and fusion of representation, blood and medical image data, characterized in that it includes a method for calculating the sum of the probability of intelligent identification of diseases of a characteristic item. The steps of intelligent recognition probability of disease corresponding to disease are as follows:

Step1. To characterize each sample, extract its corresponding organ;

Step2. Extract its organ image data, blood cell image data and blood data in the sample;

Step3. If it is a blood abnormal feature item, increase the count value of the feature item. Increase the count value of the total feature item;

Step4. For each type of organ, for the blood cell image of each organ,

If it is an abnormal feature item of blood cell image, increase the count value of this feature item,

Increase the count value of the total feature item;

Step5. For each type of organ,

For each organ tissue image,

If it is an abnormal feature item of the organ tissue image, increase the count value of the feature item,

Increase the count value of the total feature item;

Step6. If the abnormal feature count value of the feature > the number of intelligent identification of diseases of the corresponding feature item, set it as the probability of intelligent identification of hard diseases;

If, the abnormal feature count value of the feature is less than the corresponding number of intelligent identification of diseases, set it as the probability of intelligent intelligent identification of soft diseases;

Step7. Return the disease intelligence recognition probability of the feature.
An optimization method for disease analysis and identification based on multi-data association and fusion of representation, blood and medical image data, characterized in that it is an optimization method for intelligent disease identification gain, and the optimization method is used to realize intelligent identification of diseases, according to S12 of claim 1 The improved genetic optimization method of the described feature item combination is as follows:

Step1. Set each division method as chromosome;

Step2. Select the feature items of each division method as genes, and the total number of feature items as the number of genes;

Step3. Variable setting, when the feature item is selected in this division method, the variable value of the selected locus is 1;

The variable value of the unselected locus is 0;

Step4. Use to select the optimal chromosome population;

Step5. In the cross parent chromosomes A and B, select the locus that needs to be exchanged, and the gene method to exchange the locus is:

"11" in a chromosome is exchanged with "01", "10" and "00" in another chromosome, and the method of exchange is not limited to this;

Step6. Gene mutation, select the locus with "1" as the gene mutation position. If the probability value of intelligent recognition of disease is the smallest, change 1 to 0;

Step7. The objective function is to minimize the gain of intelligent disease identification.