CN108198621B - Database data comprehensive diagnosis and treatment decision method based on neural network - Google Patents
Database data comprehensive diagnosis and treatment decision method based on neural network Download PDFInfo
- Publication number
- CN108198621B CN108198621B CN201810047166.7A CN201810047166A CN108198621B CN 108198621 B CN108198621 B CN 108198621B CN 201810047166 A CN201810047166 A CN 201810047166A CN 108198621 B CN108198621 B CN 108198621B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- database
- diagnosis
- tensor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention provides a database data comprehensive diagnosis and treatment decision method based on a neural network, which uses a nasopharyngeal carcinoma database, wherein the nasopharyngeal carcinoma database comprises a large clinical database and a gene database; the clinical big database records first data formed by clinical alphanumeric information and second data extracted and identified according to clinical image information; the first data, the second data and/or the genetic data in the nasopharyngeal carcinoma database are accompanied by their temporal attributes; the method makes a diagnosis decision based on a fifth model with the support of data in the nasopharyngeal carcinoma database; the fifth model takes a plurality of the first data, the second data and the gene data as input and provides diagnosis and treatment decisions for the user based on the neural network. The invention can provide accurate intelligent diagnosis decision service based on artificial intelligence for users, and has wide application prospect.
Description
Technical Field
The invention relates to the field of medical treatment, in particular to a database data comprehensive diagnosis and treatment decision method based on a neural network.
Background
Artificial intelligence is a technical science that studies and develops theories, methods, techniques and application systems for simulating, extending and expanding human intelligence. Artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence, a field of research that includes robotics, language recognition, image recognition, natural language processing, and expert systems, among others. Artificial intelligence medicine is its expanding application in the medical field. At present, artificial intelligent medical treatment mainly comprises aspects such as virtual assistants, medical big data, medical images, intelligent voice, body-building biotechnology, health life style management, medical search, early cancer screening, artificial intelligent chips and the like. The existing medical related intelligent systems comprise an IBM company Watson robot, a CC-Cruiser congenital cataract artificial intelligence platform of Zhongshan ophthalmic center and the like.
However, the existing artificial intelligent medical system has no intelligent system in the special medical field such as nasopharyngeal carcinoma, for example, the Watson robot has relatively complete functions and can cover various common tumors, but the system has the defects of being deep in the special cancer field with common characteristics in China. Except for the Watson robot, the functions of systems such as a plurality of domestic functional platforms are single, and the effects of systematic auxiliary diagnosis and treatment cannot be achieved mainly aiming at a certain specific small-range disease or examination project.
Nasopharyngeal carcinoma is a local characteristic cancer species in Guangdong province, recurrence and metastasis are main causes of death, accurate prediction of recurrence and metastasis and guidance of individualized treatment are the keys for improving curative effect, and how to carry out accurate decision based on intelligent data analysis is a problem to be solved at present.
Disclosure of Invention
In order to solve the problems, the invention provides a database data comprehensive diagnosis and treatment decision method based on a neural network.
The invention is realized by the following technical scheme:
a neural network-based database data comprehensive diagnosis and treatment decision method uses a nasopharyngeal carcinoma database, wherein the nasopharyngeal carcinoma database comprises a large clinical database and a gene database; the clinical big database records first data formed by clinical alphanumeric information and second data extracted and identified according to clinical image information; the first data, the second data and/or the genetic data in the nasopharyngeal carcinoma database are accompanied by their temporal attributes;
the method makes a diagnosis decision based on a fifth model with the support of data in the nasopharyngeal carcinoma database; the fifth model takes a plurality of the first data, the second data and the gene data as input and provides diagnosis and treatment decisions for the user based on the neural network.
The diagnosis and treatment decision making method further comprises diagnosis and treatment decision making based on a first model, a second model, a third model and/or a fourth model, wherein the first model takes first data as input, the second model takes second data as input, the third model takes gene data as input, and the fourth model takes multiple groups of mathematical data in a nasopharyngeal carcinoma database as input.
Further, the method for constructing the fifth model includes:
extracting all or part of characteristics in different omics data;
performing high-dimensional abstraction on the associated characterization features in different omics data to obtain high-dimensional features;
completing the filtration of redundant noise characteristics;
all the high-dimensional features are connected through the full connection layer, and therefore a fifth model is generated.
Further, adjusting parameters in the fifth model by an error back propagation algorithm in combination with a random gradient descent is also included.
Further, the method further comprises the step of limiting the adjustment range of the parameters in the fifth model by utilizing the dropout, the leave relu activation function and the norm constraint.
Further, the number of the last fully connected layers and related parameters of the fifth model are adjusted by the predicted performance of the fifth model in the training set and the verification set.
Further, the training set comprises first data, second data and/or genetic data.
The invention has the beneficial effects that:
the invention provides a database data comprehensive diagnosis and treatment decision method based on a neural network, which can provide accurate intelligent diagnosis and decision service based on artificial intelligence for users and has wide application prospect.
Drawings
FIG. 1 is a schematic diagram of a nasopharyngeal carcinoma database provided in accordance with an embodiment of the present invention;
FIG. 2 is a flow chart of a method for forming a first model according to an embodiment of the present invention;
FIG. 3 is a flowchart of a method for constructing a second model according to an embodiment of the present invention;
FIG. 4 is a flowchart of a method for constructing a third model according to an embodiment of the present invention;
FIG. 5 is a flowchart of a fourth model construction method provided by an embodiment of the present invention;
FIG. 6 is a flowchart of a method for performing an analysis of multidimensional data according to an embodiment of the present invention;
FIG. 7 is a flowchart of a method for constructing a fifth model according to an embodiment of the present invention;
FIG. 8 is a schematic diagram of a cloud system provided by an embodiment of the invention;
FIG. 9 is a schematic diagram of an analysis decision server provided by an embodiment of the present invention;
fig. 10 is a schematic diagram of a diagnosis and treatment decision terminal according to an embodiment of the present invention;
fig. 11 is a flowchart of a method for using in an online auxiliary diagnosis and treatment system according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings.
The embodiment of the invention provides an artificial intelligent auxiliary diagnosis and treatment decision system for nasopharyngeal carcinoma. Specifically, the design of the composition structure of the nasopharyngeal carcinoma database in the embodiment of the present invention is based on the following consideration for the purpose of research:
the method deeply researches the multi-scale heterogeneous association relationship among the macro multi-modal images, the clinical pathological phenotype and the micro genotype so as to establish the linkage relationship between the clinical phenotype and the micro genotype, thereby completing the multi-scale, multi-modal and big data description of various states of the nasopharyngeal darcinoma from the macro to the micro, and establishing a prediction model which can accurately and comprehensively predict the occurrence of the distant metastasis and recurrence of the nasopharyngeal darcinoma.
Collecting clinical data, image data and gene data of patients with nasopharyngeal carcinoma, establishing a large-scale multi-source heterogeneous multimathematic database, and finally developing a large-data cloud diagnosis and treatment analysis system for nasopharyngeal carcinoma and used for assisting in guiding accurate treatment.
In view of the above, the nasopharyngeal carcinoma database provided in the embodiment of the present invention is shown in fig. 1, and includes a clinical big database and a gene database.
Specifically, the clinical big database records first data formed by clinical alphanumeric information and second data extracted and identified according to clinical image information. The first data can be in a text and/or digital form, the second data can be in a picture and/or video form and is stored in a digital, binary large object, link or multimedia form, and the first data and the second data can be used as a data source of multi-dimensional analyzable data for subsequent diagnosis and treatment decisions.
In particular, the large clinical database may include an alphanumeric database for storing first data corresponding to clinical alphanumeric information including, but not limited to, exam findings, treatment data, and treatment status data. Specifically, the examination and inspection report comprises medical record data, inspection indexes, and examination and inspection reports of pathology, ultrasound, image and/or nuclear medicine; the treatment data includes, but is not limited to, radiotherapy, chemotherapy, surgery, targeting, biological treatment, and/or a traditional chinese treatment, and the treatment data includes, but is not limited to, follow-up prognosis data, and treatment effect and complication data.
Specifically, the large clinical database may further include an image database, where the image database stores second data, and the corresponding clinical image information includes, but is not limited to, various clinical images such as images, radiotherapy target regions, ultrasound, nuclear medicine, and pathology.
It is emphasized that the first data, the second data and/or the genomics data in the nasopharyngeal carcinoma database are recorded along with the time attributes thereof, and the time attributes have a certain role in various diagnosis and treatment decisions, and by taking the index time attributes into consideration of the diagnosis and treatment decisions, a time-dependent decision result can be provided for the survival prognosis of the patient.
With the data support of the nasopharyngeal carcinoma database, the diagnosis and treatment decision system can analyze, predict and guide the nasopharyngeal carcinoma based on a plurality of diagnosis and treatment decision models.
In one possible embodiment, the medical decision system can include a first medical module that can provide nasopharyngeal carcinoma analysis, prediction, and decision guidance based on a first model. The first model takes first data as a data source, and the forming method of the first model is shown in fig. 2 and comprises the following steps:
s1, analyzing the first data, correlating all factors in the first data with prognosis and treatment effect of a patient, and removing factors with strong correlation and high collinearity.
S2, screening out characteristic factors which have definite and obvious influence on the prognosis of the patient, and establishing a first model.
And S3, carrying out verification analysis on the first model.
The first model may be input as the first data or some field or fields in the first data and output as prognostic indicators for the patient, including but not limited to: n-year survival, and tumor recurrence, metastasis and/or complications; prognostic indicators are used to indicate the final outcome of the patient and the trend of the development of the nasopharyngeal carcinoma. Further, an index having directivity for patient prognosis and a weight corresponding to the index may be obtained in the first data in the first model.
Further, with the continuous expansion of the data volume of the large clinical database and the development of medical shared data in various regions, data from the data center or different data centers or medical systems can be introduced into the first model obtained in S2 to be verified and analyzed, and relevant parameters can be corrected. The content of a large clinical database is continuously enlarged, and meanwhile, the relevant parameters of the first model can be corrected in real time, so that the first model is continuously perfected to reach the optimum, and a more valuable reference is provided for nasopharyngeal carcinoma analysis, prediction and decision guidance.
The first model is established by a first data bit research object, the first data acquisition way is simple in clinical work, and certain advantages are achieved in the aspect of data acquisition, so that the first model has a good application prospect in clinical work.
In another possible embodiment, the clinical decision system may further include a second clinical module that may provide nasopharyngeal carcinoma analysis, prediction, and decision guidance based on a second model. The second model takes the second data as a data source. The second model construction method is shown in fig. 3, and includes:
and S10, preprocessing the image data.
And S20, segmenting and three-dimensionally reconstructing the image.
And S30, carrying out high-throughput quantitative analysis on the image, and constructing a second model based on the image.
The second model may be input with the second data or some portion of the second data and output with prognostic indicators for the patient, including but not limited to: n-year survival, and tumor recurrence, metastasis and/or complications; prognostic indicators are used to indicate the final outcome of the patient and the trend of the development of the nasopharyngeal carcinoma. Further, an index having directivity for patient prognosis and a weight corresponding to the index may be obtained in the second data in the second model.
Specifically, the image can be preferably a nasopharyngeal carcinoma CT/MRI image, and after the image is segmented and three-dimensionally reconstructed, high-throughput quantitative analysis of the digitized multi-parameter CT/MRI and other images can be carried out. Specific analysis contents include but are not limited to characteristic parameter data extraction, characteristic parameter data correlation analysis, clustering and statistical characteristic exploration, high-sensitivity transfer recurrence image parameters are searched from a large number of multi-modal medical images of recurrence transfer nasopharyngeal carcinoma, and a second model based on images of nasopharyngeal carcinoma CT/MRI and the like is constructed by combining prognosis conditions of recurrence, transfer and the like of a patient.
Further, with the continuous expansion of the data volume of the large clinical database and the development of medical shared data in various regions, data from the data center or different data centers or medical systems can be introduced into the second model obtained in S30 to be verified and analyzed, and relevant parameters can be corrected. Furthermore, while the content of the large clinical database is continuously enlarged, the relevant parameters of the second model can be corrected in real time, so that the model is continuously improved to reach the optimum.
The second model is a model taking the second data as a research object, along with improvement of medical conditions and popularization of imaging technology, the effect of the second data in clinical work is increasingly prominent, and the directionality of the second data on medical diagnosis is gradually enhanced, so that the second model also has a better application prospect in clinical work.
Further, the gene database is used for storing gene data, compared with the first data and the second data, the gene data is more difficult to obtain and analyze, but the effect of the gene database on the prognosis analysis of nasopharyngeal carcinoma is also very prominent. The establishment of the gene database can create conditions for the following work: the whole genome sequencing data of recurrent metastatic nasopharyngeal carcinoma is analyzed at a microscopic level, genotype variation in the processes of tumor occurrence, development and apoptosis is explored, and potential metastatic recurrent driving genes and molecular markers are discovered in order to provide an important basis for clinical accurate medical research.
Therefore, the nasopharyngeal carcinoma database in the embodiment of the invention is a multi-source heterogeneous multigroup chemical database integrating a large clinical database and a gene database, and the stored data form multidimensional analyzable data.
In another possible embodiment, the clinical decision system may further include a third clinical module that may provide nasopharyngeal carcinoma analysis, prediction, and decision guidance based on a third model. The third model uses gene data as a data source. The method for constructing the third model is shown in fig. 4, and includes:
s100, carrying out whole genome sequencing on the matched primary focus, metastatic focus and recurrent focus of the nasopharyngeal carcinoma.
S200, performing bioinformatics analysis on the sequenced data.
And (3) carrying out comprehensive bioinformatics analysis on the sequenced data, finding highly-related gene targets, differential molecules and molecular markers in nasopharyngeal carcinoma metastasis and recurrence samples, and finding meaningful metastasis and recurrence driving genes and mutation information and metastasis and recurrence related pathways.
S300, constructing a third model based on nasopharyngeal darcinoma gene data according to the analysis result.
And (3) carrying out functional verification on the transfer driving gene and the related mutation information obtained by analysis by adopting in-vivo and in-vitro experiments, and constructing a third model based on nasopharyngeal carcinoma gene data. The third model may be input with all or part of the genetic data and output with prognostic indicators for the patient, including but not limited to: n-year survival, and tumor recurrence, metastasis and/or complications; prognostic indicators are used to indicate the final outcome of the patient and the trend of the development of the nasopharyngeal carcinoma. Further, an index having directivity for patient prognosis and a weight corresponding to the index may be obtained in the gene data in the third model.
Further, with the continuous expansion of the data volume of the gene database and the development of medical shared data across the country, data from different data centers or medical systems can be introduced into the third model obtained in step S300 to be verified and analyzed, and relevant parameters can be corrected. Furthermore, the content of the gene database is continuously enlarged, and meanwhile, the relevant parameters of the third model can be corrected in real time, so that the model is continuously perfected to achieve the optimization.
The third model is a model established by taking gene data as a research object, the gene data has increasingly prominent effect in clinical work along with popularization of gene technology and related research, and the directionality of the gene data on medical diagnosis is gradually enhanced, so that the third model also has a better application prospect in clinical work.
The data sources of the first model, the second model and the third model are different, and the diagnosis and treatment decision for nasopharyngeal carcinoma based on multiple data sources can take various data sources into consideration, so that more comprehensive diagnosis and treatment guidance is provided for patients. Based on this recognition, in another possible embodiment, the clinical decision system can further include a fourth treatment module that can provide nasopharyngeal carcinoma analysis, prediction, and decision guidance based on a fourth model. The fourth model can take a plurality of data in the first data, the second data and the third data in the nasopharyngeal darcinoma database as research objects to guide diagnosis and treatment decision for users. The fourth model construction method is shown in fig. 5, and includes:
and P1, performing integrated analysis on the multiple groups of chemical data, and establishing a data storage model after quality evaluation.
Specifically, the plurality of sets of genomic data may be a plurality of first data, second data, and genetic data, wherein the first data, the second data, and the genetic data are omic data.
And P2, comprehensively and systematically exploring the relevance among multiple groups of mathematical data, and establishing a fourth model based on the integration of the multiomic data.
The fourth model in the embodiment of the invention comprises a public discovery model, and the public discovery model can be specifically constructed based on a sparse expression theory.
In another possible implementation, in the execution of P2, a feature model facing tensor description can be proposed in the framework of tensor expression and matrix expression; establishing a tensor pairing energy equation based on prior knowledge constraint; analyzing the internal relation between the characteristic tensor description method, the prior knowledge constraint design, the tensor pairing method and the local public substructure; testing the robustness of the model aiming at the data with high noise pollution; aiming at mass data, an efficient constraint problem decomposition mechanism is established to quickly optimize and solve, so that a public discovery model is finally obtained.
Specifically, the embodiment of the present invention provides a method for performing integrated analysis on multiple sets of mathematical data, as shown in fig. 6, including:
firstly, a matrix decomposition model facing multiple groups of mathematical data is established, multi-dimensional data are decomposed into the same base space, and the interaction relation between coefficient matrixes is analyzed, so that the incidence relation between the multiple groups of mathematical data is found.
Specifically, an association rule between the research genotype and the phenotype is established, and dynamic association between the important gene target point and factors such as the image phenotype and the tumor stage grading is found.
Secondly, a tensor decomposition model facing multiple groups of mathematical data is established, the multiple groups of mathematical data are decomposed into three base spaces by adopting a method of three implicit variable base spaces, a central kernel tensor independent of the multiple groups of mathematical data is extracted, and the common relation among the kernel tensors is researched so as to find out the common relation.
Specifically, the decomposition characteristics of an implicit variable base space under different constraints are researched, and an association function oriented to a multi-constraint central core tensor is designed.
Thirdly, establishing a tensor matching model based on tensor description, discussing a tensor feature description method of single-component chemical data, and researching the influence of tensor feature dimension and quantitative description on tensor matching; designing a pairing method of feature tensors, and constructing a tensor pairing model based on prior knowledge constraint.
And finally, establishing a common module discovery model aiming at the analysis result.
Based on the research method of the public discovery model, the embodiment of the invention provides an innovative integrated model optimization function, which is the core content of the public discovery model and is used for selecting an index with directivity for the prognosis of a nasopharyngeal carcinoma patient from multiple groups of mathematical data:
Subjectto:Xi,j∈{0,1}
XI1≤b1
XTI2≤b2
wherein H represents omics data S1And S2The similarity between them, X denotes an allocation matrix,
representing data S1The correlation relationship of the internal characteristics is shown,
representing data S2The incidence relation of the internal features, the first item of the formula represents that the distribution matrix X meets the distribution of the similarity matrix H as much as possible, and the second item and the third item respectively represent that the data S is blended in1And data S2The prior knowledge improves the accuracy, and the fourth item represents that the distribution matrix has sparsity, so that the data S can be better explained1And data S2The association relationship between them.
Wherein L isMTo normalize the Laplace matrix, it is expressed as follows
From the Laplace matrix properties, the following equations can be derived
Finally, solving the optimization function through a gradient descent method to obtain a data integration result, as shown in the following algorithm 1
Algorithm 1
Inputting: m1,M2,WM1,WM2
1: initializing lambda1,λ2,β,X(0),γ(1),M,ρ>1,δ∈(0,1),ε>0,k＝1
2: normalization of M1,M2
3: updating distribution matrix X by using gradient descent method to solve following formula(k)
4: setting M as rho M if
5: updating gamma(k+1)＝min(γ(k)-Mg(X(k)),0)
6: setting k as k +1
7: repeating the steps 3,4,5 and 6 until | g (X)(k))||<ε
And (3) outputting: x
Furthermore, on the basis of obtaining a directional index for prognosis based on the common discovery model, in order to improve the comprehensive diagnosis and treatment effect, the fourth model may also take into account the weight of each directional index in prognosis determination. There are many methods for obtaining the weight based on the fourth model, and the embodiment of the present invention summarizes three possible implementations.
In one possible implementation, the output of the fourth model may be simply expressed as P ═ K1a1+ K2a2+ K3A3+ … … + KmAm. P is the probability of occurrence of various survival possibilities such as no disease progression within 5 years/5 years under the conditions of A1+ A2+ A3+ … … + Am. And obtaining the value of A1 … … Am according to the data in the nasopharyngeal darcinoma database, thus obtaining the weight of various directional indexes in prognosis judgment.
In another possible embodiment, the output P of the fourth model may be expressed as the result of a combination of multiple indicators, including simultaneous, staged, and mixed effects of directional indicators for each prognosis. In the expression method, the weight of various directional indexes in prognosis judgment is obtained by analyzing data in a nasopharyngeal carcinoma database.
In another possible embodiment, the weights of the various directivity indexes in the prognosis determination may be obtained based on the research results of the first model, the second model and/or the third model. In this embodiment, the construction of the fourth model comprises the steps of:
selecting an index with directivity for prognosis according to the public discovery model;
judging whether the index belongs to the first data, the second data or the gene data;
if the index belongs to the first data, acquiring the weight of the index according to a first model;
if the index belongs to the second data, acquiring the weight of the index according to a second model;
if the index belongs to gene data, acquiring the weight of the index according to a third model;
and after the weights of all directional indexes are obtained, diagnosis and treatment decision guidance is carried out according to the indexes and the weighting results of the weights.
Obviously, in the actual use process of the diagnosis and treatment decision system, the feedback of doctors and patients needs to be continuously acquired, and the common module discovery model, the first model, the second model and the third model are continuously corrected by combining continuously updated data and documents, so that more power is provided for artificial intelligent auxiliary diagnosis and treatment.
In order to obtain more comprehensive diagnosis and treatment results based on the nasopharyngeal carcinoma database, in another possible embodiment, the diagnosis and treatment decision system can further comprise a fifth diagnosis and treatment module, and the fifth diagnosis and treatment module can provide nasopharyngeal carcinoma analysis, prediction and decision guidance based on a fifth model. The fifth model can also take multiple groups of mathematical data in the nasopharyngeal darcinoma database as data sources, and is different from the construction method of the fourth model, and the fifth model provides diagnosis and treatment decisions for users based on a neural network. The method for constructing the fifth model is shown in fig. 7, and includes:
p10, extracting features of the different omics data that are valuable for the generation of the fifth model.
The specific extraction method may be various, and the feature extraction may also be performed with reference to the common discovery model in the fourth model.
And P20, performing high-dimensional abstraction on the associated characteristic features in different omics data to obtain high-dimensional features.
P30, filtering of redundant noise features is accomplished.
In the three steps, the feature expression from the bottom layer to the high layer can be extracted from the first data and the gene data by adopting a convolutional neural network; and mining useful characteristics of the second data by adopting an integration method, and filtering irrelevant interference characteristics by combining different norm constraints through traditional logistic regression.
Performing high-dimensional abstraction on the associated characteristic features in different omics data to obtain high-dimensional features, and finishing filtering redundant noise features; thereby creating conditions for P40.
P40, concatenating all the high-dimensional features through the full concatenation layer, thereby generating a fifth model.
Specifically, all the high-dimensional features can be connected through the full connection layer to generate a diagnosis and treatment decision model, and when the newly collected data volume for training the model reaches a specified magnitude, model training is restarted to optimize model parameters, so that the model continuously learns and adjusts and perfects by itself, and a fifth model is formed.
Specifically, the fifth model according to the embodiment of the present invention may be implemented by establishing a convolutional neural network and learning the convolutional neural network.
In the construction process of the fifth model, data in the nasopharyngeal darcinoma database are integrated through a full-connection layer, parameters are adjusted through an error back propagation algorithm in combination with random gradient descent, the adjustment range of the parameters is limited by utilizing a dropout, a leave relu activation function and norm constraints, the sparsity of characteristics is guaranteed, the over-fitting condition is avoided, and therefore the efficiency of the fifth model is further guaranteed. The number of the last full connection layers of the model and related parameters are adjusted by the prediction performance of the fifth model in the training set and the verification set, so as to provide valuable diagnosis opinions for doctors.
The fourth model and the fifth model are models of research objects with multiple groups of science data, the content of a nasopharyngeal carcinoma database can be comprehensively used so as to obtain a more comprehensive decision result, the first model, the second model and the third model are models with single group of science data as research objects, diagnosis and treatment can be carried out on the basis of certain group of science data, and in a preferred implementation mode, the first model, the second model, the third model, the fourth model and the fifth model can exist in parallel or preferentially exist in one or more types, so that a higher-quality diagnosis and treatment decision service is provided for a user from multiple angles.
On the basis of the above content, the embodiment of the present invention further provides a logic architecture of the artificial intelligence auxiliary diagnosis and treatment decision system for nasopharyngeal carcinoma, wherein the artificial intelligence auxiliary diagnosis and treatment decision system for nasopharyngeal carcinoma may specifically be an online diagnosis and treatment system, specifically, the online diagnosis and treatment system may be a cloud system, and a data supporter (nasopharyngeal carcinoma database) as the online diagnosis and treatment system may also be arranged in the cloud, so that obviously, cloud storage is used to facilitate data integration and diagnosis and treatment analysis.
As shown in fig. 8, the diagnosis and treatment decision system may include a diagnosis and treatment decision server disposed at a cloud end and a diagnosis and treatment decision terminal in communication connection with the diagnosis and treatment decision server, where the diagnosis and treatment decision system may specifically use a B-S architecture or a C-S architecture, and the diagnosis and treatment decision terminal responds to a user instruction and sends a data request corresponding to the user instruction to the diagnosis and treatment decision server cluster; and the diagnosis and treatment decision server responds to the data request, generates a data response and transmits the data response to the diagnosis and treatment decision terminal, so that the diagnosis and treatment decision terminal can display a result corresponding to the user instruction to a user.
Further, in order to provide a diagnosis and treatment service for a user comprehensively, the diagnosis and treatment decision server may specifically be a server cluster (a diagnosis and treatment decision server cluster), the diagnosis and treatment decision server cluster includes a user interaction server, an analysis decision server and a data server, the user interaction server is used for performing data interaction with the diagnosis and treatment decision terminal and performing user management, the analysis decision server is used for providing the diagnosis and treatment decision service for the user, the data server is used for performing data processing, and a nasopharyngeal carcinoma database is arranged for performing interaction. The analysis decision server, as shown in fig. 9, preferably includes a first diagnosis module, a second diagnosis module, a third diagnosis module, a fourth diagnosis module and a fifth diagnosis module, and the functions of the first diagnosis module, the second diagnosis module, the third diagnosis module, the fourth diagnosis module and the fifth diagnosis module are as described above, and it is not repeated here.
Furthermore, the diagnosis and treatment decision terminal is provided with functions of user registration, data management, query, annotation, comparison, retrieval, display and the like. Specifically, as shown in fig. 10, the diagnosis and treatment decision terminal includes:
and the user registration module is used for registering the diagnosis and treatment decision server cluster so that the user can conveniently use various services provided by the diagnosis and treatment decision server cluster.
And the data management module is used for managing personal data of the user.
And the query module is used for querying the target data from the diagnosis and treatment decision server cluster by the user.
And the annotation module is used for marking on the display interface by the user.
And the comparison module is used for comparing the similar data and identifying the comparison result so as to facilitate the user to quickly find the difference between the similar data.
And the retrieval module is used for quickly searching the target data by the user.
And the display module is used for carrying out data visualization.
The diagnosis and treatment decision terminal can be any intelligent device loaded with software interacting with the diagnosis and treatment decision server, such as a PC, a mobile phone or a tablet and the like.
The diagnosis and treatment decision terminal also comprises an interface switching module, and the interface switching module is used for providing different display interfaces for the user according to different login identities of the user. Specifically, the interface switching module additionally provides different functions according to different identities of users. If the user identity is the patient user, functions of accessing data, checking diagnosis results, treating schemes and the like are provided for the user; if the user identity is a doctor user, the user provides functions of accessing data, making a diagnosis decision, performing prognosis evaluation and the like.
Specifically, in the B-S framework, diagnosis and treatment results of cloud analysis can be visualized by adopting a simple and friendly operation interface at the front end of a website. The visualization can be specifically realized by a visualization module of the diagnosis and treatment decision terminal, and the visualization module comprises a data early warning unit, a chart linkage unit and a comparison and splitting unit. The method is convenient for the operation of the clinical professional doctors, and provides a clear medical image display interface and a clear case data display interface so that the clinical professional doctors can make evaluation and diagnosis better.
The online diagnosis and treatment system not only provides primary prediction results of nasopharyngeal carcinoma recurrence and metastasis staging, but also provides functions of managing, annotating, comparing, retrieving and displaying magnitude omics data and clinical information above PB. The online diagnosis and treatment system can provide direction guidance for gene targets, image tumor regions and accurate individualized medical treatment, provide a visual display platform for disease diagnosis for doctors and patients, provide a reliable and effective model system for tumor research and treatment, and provide a reference blue book for other tumor researches.
Specifically, on the basis of the above contents, an embodiment of the present invention provides an online diagnosis and treatment system using method as a diagnosis and treatment example, as shown in fig. 11, including:
s101, acquiring clinical alphanumeric information, clinical image information and/or gene data, and incorporating the data into a nasopharyngeal carcinoma database by combining time attributes of the data.
In particular, the database table structure may be as shown in Table 1:
TABLE 1
S102, selecting a diagnosis and treatment decision model capable of being used for decision making, and using the diagnosis and treatment decision model for survival prognosis, complication and analysis.
Specifically, the diagnosis and treatment decision model for decision making may be one or more of a first model, a second model, a third model, a fourth model and a fifth model.
In this step, the model used for decision making may be a fourth model, and the output of the fourth model is a weighted result of an index having directivity for prognosis, and may be briefly expressed as P ═ K1a1+ K2a2+ K3A3+ … … + KmAm. P is the probability of occurrence of various survival possibilities such as no disease progression within 5 years/5 years under the conditions of A1+ A2+ A3+ … … + Am. The larger the P, the longer the patient was obtained under A1+ A2+ A3+ … … + Am conditions. A histogram model or the like can be established for predicting survival data such as the probability of a1+ a2+ A3+ … … + Am for survival of multiple juveniles.
And the directional adjustment can be carried out on Ai according to the sizes of the coefficients Ki of different items Ai so as to achieve better survival benefit.
S103, according to the existing model, combining the current case, obtaining a diagnosis and treatment decision reference scheme and obtaining follow-up data.
And S104, continuously learning and optimizing the existing diagnosis and treatment decision model with the addition of new cases.
In the specific implementation process of the embodiment of the invention, the following results can be obtained:
(1) patient side:
after logging in, the patient fills in the basic information of the patient, inputs the relevant examination and test results, and can obtain relevant contents such as relevant disease diagnosis, prognosis estimation, attention matters during treatment and the like. And as treatment progresses, different recommendations are made based on their different inputs.
(2) Aspects of the investigator
And establishing relevant models such as treatment suggestion, prognosis and the like according to the existing database. And according to the input information of the patient, new big data is formed, and the relevant parameters of the model are corrected to obtain the model which can reflect the real situation most.
And integrating data information of clinical data and gene data to discover a macro-phenotype-a marker highly related to the micro-genotype at multiple levels, so as to establish the relationship from a gene target to an image phenotype and further establish an artificial intelligent precise medical system based on big data.
In addition, the following implementation steps in the embodiment of the invention are all proposed for the first time in the related field:
(1) and constructing a multigroup chemical database with time attributes, and establishing a prediction model of multiomic prognosis and complications and the like according to time.
(2) Searching a high-sensitivity metastasis and recurrence image mark from multi-modal medical images of mass recurrence and metastasis nasopharyngeal carcinoma;
(3) highly relevant gene targets, differential molecules and molecular markers are found in nasopharyngeal carcinoma metastasis and recurrence samples by using a genomics biological analysis technology;
(4) integrating multi-level heterogeneous data such as clinical alphanumeric information, clinical image information and gene data to discover a marker highly associated under macro phenotype-observation genotype multi-level, thereby establishing association from a gene target to an image phenotype;
(5) the online diagnosis and treatment system integrates management, annotation, comparison, retrieval, omics data display, recurrence and metastasis staging and prediction prognosis.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it is therefore to be understood that the invention is not limited by the scope of the appended claims.
Claims (7)
1. A database data comprehensive diagnosis and treatment decision method based on a neural network is characterized in that a nasopharyngeal carcinoma database is used, and the nasopharyngeal carcinoma database comprises a large clinical database and a gene database; the clinical big database records first data formed by clinical alphanumeric information and second data extracted and identified according to clinical image information; the first data, the second data and/or the genetic data in the nasopharyngeal carcinoma database are accompanied by their temporal attributes;
the method makes a diagnosis decision based on a fifth model with the support of data in the nasopharyngeal carcinoma database; the fifth model takes a plurality of the first data, the second data and the gene data as input and provides diagnosis and treatment decisions for the user based on the neural network;
the method further comprises the following steps:
performing integration analysis on multiple groups of chemical data, and establishing a data storage model after quality evaluation, wherein the multiple groups of chemical data are multiple types of first data, second data and gene data, and the first data, the second data and the gene data are all omic data;
comprehensively and systematically exploring the relevance among multiple groups of academic data, and establishing a fourth model based on the integration of the multiple groups of academic data, wherein the fourth model comprises a public discovery model, and the public discovery model is obtained by the following method:
under the framework of tensor expression and matrix expression, a feature model facing tensor description is provided; establishing a tensor pairing energy equation based on prior knowledge constraint; analyzing the internal relation between the characteristic tensor description method, the prior knowledge constraint design, the tensor pairing method and the local public substructure; testing the robustness of the model; aiming at mass data, establishing a constraint problem decomposition mechanism, and carrying out optimized solution to obtain a public discovery model;
the omics data are integrated by the following method:
s1, establishing a matrix decomposition model facing multiple groups of mathematical data, decomposing multi-dimensional data into the same base space, analyzing the interaction relation between coefficient matrixes, and finding dynamic association between a gene target and an image phenotype and tumor stage grading;
s2, establishing a tensor decomposition model facing multiple groups of mathematical data, decomposing the multiple groups of mathematical data into three base spaces by adopting a method of three implicit variable base spaces, extracting a central core tensor independent of the multiple groups of mathematical data, researching the common relation among the core tensors to find the common relation of the core tensors, researching the decomposition characteristics of the implicit variable base spaces under different constraints, and designing a correlation function facing the multi-constraint central core tensor;
s3, establishing a tensor matching model based on tensor description, designing a tensor feature description method of the monamics data, and researching the influence of tensor feature dimension and quantitative description on tensor matching; designing a pairing method of feature tensors, and constructing a tensor pairing model based on prior knowledge constraint;
s4, establishing a common module discovery model aiming at the analysis result, and integrating a model optimization function, wherein the integrated model optimization function is the core content of the common discovery model and is used for selecting an index with directivity for the prognosis of the nasopharyngeal carcinoma patient from multiple groups of mathematical data:
Subjectto:Xij∈{0,1}
XI1≤b1
XTI2≤b2，
wherein H represents omics data S1And S2The similarity between them, X denotes an allocation matrix,
representing data S1The correlation relationship of the internal characteristics is shown,
representing data S2The incidence relation of the internal features, the first item of the formula represents that the distribution matrix X meets the distribution of the similarity matrix H, and the second item and the third item respectively represent that the data S is blended in1And data S2The fourth term represents the sparsity of the allocation matrix, thereby interpreting the data S1And data S2Correlation between, LMIn order to normalize the laplace matrix, the matrix,
according to the property of Laplace matrix, the following formula is obtained
And solving the optimization function to obtain a data integration result.
2. The method according to claim 1, further comprising making a diagnosis decision based on a first model, a second model, a third model and/or a fourth model, wherein the first model takes first data as input, the second model takes second data as input, the third model takes genetic data as input, and the fourth model takes multimathematical data in the nasopharyngeal carcinoma database as input.
3. The database data comprehensive diagnosis and treatment decision method based on the neural network as claimed in claim 2, wherein: the method for constructing the fifth model comprises the following steps:
extracting feature expression from a bottom layer to a high layer from the first data and the gene data by adopting a convolutional neural network;
performing feature mining on the second data by adopting an integration method, and filtering interference features by combining logistic regression with norm constraint;
performing high-dimensional abstraction on the associated characterization features in different omics data to obtain high-dimensional features;
connecting the obtained high-dimensional features through a full-connection layer to generate a diagnosis and treatment decision model;
and when the number of the training sets reaches a preset value, restarting the training of the diagnosis and treatment decision model, so that the model optimizes the relevant parameters through self-learning, thereby forming a fifth model.
4. The database data comprehensive diagnosis and treatment decision method based on the neural network as claimed in claim 3, wherein: further comprising adjusting parameters in the fifth model by an error back-propagation algorithm in combination with a random gradient descent.
5. The database data comprehensive diagnosis and treatment decision method based on the neural network as claimed in claim 3, wherein: further comprising limiting the adjustment range of the parameters in the fifth model using dropout, leakyrelu activation functions and norm constraints.
6. The database data comprehensive diagnosis and treatment decision method based on the neural network as claimed in claim 5, wherein:
and adjusting the number of the last full connection layers and related parameters of the fifth model through the predicted performances of the fifth model in the training set and the verification set.
7. The database data comprehensive diagnosis and treatment decision method based on the neural network as claimed in claim 6, wherein:
the training set includes first data, second data, and/or genetic data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810047166.7A CN108198621B (en) | 2018-01-18 | 2018-01-18 | Database data comprehensive diagnosis and treatment decision method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810047166.7A CN108198621B (en) | 2018-01-18 | 2018-01-18 | Database data comprehensive diagnosis and treatment decision method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108198621A CN108198621A (en) | 2018-06-22 |
| CN108198621B true CN108198621B (en) | 2022-03-08 |
Family
ID=62590125
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810047166.7A Active CN108198621B (en) | 2018-01-18 | 2018-01-18 | Database data comprehensive diagnosis and treatment decision method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108198621B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102494A (en) * | 2018-07-04 | 2018-12-28 | 中山大学中山眼科中心 | A kind of After Cataract image analysis method and device |
| CN109087702B (en) * | 2018-08-03 | 2021-07-16 | 厦门大学 | Four-diagnosis representation information fusion method for traditional Chinese medicine health state analysis |
| CN109182526A (en) * | 2018-10-10 | 2019-01-11 | 杭州翱锐生物科技有限公司 | Kit and its detection method for early liver cancer auxiliary diagnosis |
| CN109903846A (en) * | 2019-02-21 | 2019-06-18 | 吉林禾熙科技开发有限公司 | Screening lung cancer system and method based on medical big data |
| CN109887589A (en) * | 2019-02-21 | 2019-06-14 | 吉林禾熙科技开发有限公司 | Artificial intelligence medical treatment big data system |
| US11154240B2 (en) | 2019-04-02 | 2021-10-26 | Kpn Innovations Llc | Methods and systems for utilizing diagnostics for informed vibrant constitutional guidance |
| CN110277165B (en) * | 2019-06-27 | 2021-06-04 | 清华大学 | Auxiliary diagnosis method, device, equipment and storage medium based on graph neural network |
| CN111128328A (en) * | 2019-10-25 | 2020-05-08 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Nasopharyngeal carcinoma structured image report and data processing system and method |
| US20210312487A1 (en) * | 2019-12-03 | 2021-10-07 | Click Therapeutics, Inc. | Apparatus, system, and method for determining demographic information to facilitate mobile application user engagement |
| CN111508618B (en) * | 2020-04-13 | 2023-09-15 | 南京巨鲨显示科技有限公司 | Manual intelligent auxiliary platform system for perioperative period |
| CN111768818A (en) * | 2020-06-05 | 2020-10-13 | 上海市浦东医院（复旦大学附属浦东医院） | Artificial intelligence-based pelvis fracture first-aid decision-making system and method |
| CN112270660B (en) * | 2020-09-30 | 2022-09-09 | 四川大学 | Nasopharyngeal carcinoma radiotherapy target area automatic segmentation method based on deep neural network |
| CN112017785B (en) * | 2020-11-02 | 2021-02-05 | 平安科技（深圳）有限公司 | Disease risk prediction system, method, device, equipment and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102827941A (en) * | 2012-09-06 | 2012-12-19 | 温州医学院附属第一医院 | Gene chip testing kit for recurrent nasopharyngeal carcinoma |
| CN107247881A (en) * | 2017-06-20 | 2017-10-13 | 北京大数医达科技有限公司 | A kind of multi-modal intelligent analysis method and system |
| CN107273685A (en) * | 2017-06-14 | 2017-10-20 | 广东工业大学 | A kind of data analysing method of multi-modal big data for clinical disease |
| CN107463774A (en) * | 2017-07-21 | 2017-12-12 | 温馨港网络信息科技（苏州）有限公司 | Analysis on the health status Forecasting Methodology and system based on big data |
-
2018
- 2018-01-18 CN CN201810047166.7A patent/CN108198621B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102827941A (en) * | 2012-09-06 | 2012-12-19 | 温州医学院附属第一医院 | Gene chip testing kit for recurrent nasopharyngeal carcinoma |
| CN107273685A (en) * | 2017-06-14 | 2017-10-20 | 广东工业大学 | A kind of data analysing method of multi-modal big data for clinical disease |
| CN107247881A (en) * | 2017-06-20 | 2017-10-13 | 北京大数医达科技有限公司 | A kind of multi-modal intelligent analysis method and system |
| CN107463774A (en) * | 2017-07-21 | 2017-12-12 | 温馨港网络信息科技（苏州）有限公司 | Analysis on the health status Forecasting Methodology and system based on big data |
Non-Patent Citations (1)
| Title |
|---|
| 应用人工神经网络方法预测鼻咽癌患者5年生存状态;华贻军，洪明晃等;《肿瘤学杂志》;20060710;300-303 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108198621A (en) | 2018-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108198621B (en) | Database data comprehensive diagnosis and treatment decision method based on neural network | |
| CN108335756B (en) | Nasopharyngeal carcinoma database and comprehensive diagnosis and treatment decision method based on database | |
| CN108206056B (en) | Nasopharyngeal darcinoma artificial intelligence assists diagnosis and treatment decision-making terminal | |
| US10275711B2 (en) | System and method for scientific information knowledge management | |
| US9141913B2 (en) | Categorization and filtering of scientific data | |
| US8364665B2 (en) | Directional expression-based scientific information knowledge management | |
| Huang et al. | Integration of patch features through self-supervised learning and transformer for survival analysis on whole slide images | |
| AU2017338775A1 (en) | Phenotype/disease specific gene ranking using curated, gene library and network based data structures | |
| Zhang et al. | Protein complex prediction in large ontology attributed protein-protein interaction networks | |
| CN108320807A (en) | A kind of nasopharyngeal carcinoma artificial intelligence assisting in diagnosis and treatment decision cloud system | |
| CN108320797B (en) | Nasopharyngeal carcinoma database and comprehensive diagnosis and treatment decision method based on database | |
| CN114003734A (en) | Breast cancer risk factor knowledge system model, knowledge map system and construction method | |
| Matos et al. | Research techniques made simple: mass cytometry analysis tools for decrypting the complexity of biological systems | |
| Heinzel et al. | Functional molecular units for guiding biomarker panel design | |
| Wang et al. | Poisson-based self-organizing feature maps and hierarchical clustering for serial analysis of gene expression data | |
| Wu et al. | A personalized diagnostic generation framework based on multi-source heterogeneous data | |
| CN108335748A (en) | A kind of nasopharyngeal carcinoma artificial intelligence assisting in diagnosis and treatment policy server cluster | |
| Sethi et al. | Long Short-Term Memory-Deep Belief Network based Gene Expression Data Analysis for Prostate Cancer Detection and Classification | |
| Gancheva et al. | Software architecture for adaptive in silico knowledge discovery and decision making based on big genomic data analytics | |
| WO2009039425A1 (en) | Directional expression-based scientific information knowledge management | |
| Gancheva | A big data management approach for computer aided breast cancer diagnostic system supporting precision medicine | |
| Wu et al. | Be-1DCNN: a neural network model for chromatin loop prediction based on bagging ensemble learning | |
| Bernasconi et al. | PoliViews: A comprehensive and modular approach to the conceptual modeling of genomic data | |
| Lengerich | Sample-Specific Models for Precision Medicine | |
| Tang et al. | Multi-Omics Data Mining Techniques: Algorithms and Software |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |