CN113707247B - Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain - Google Patents
Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain Download PDFInfo
- Publication number
- CN113707247B CN113707247B CN202111012695.1A CN202111012695A CN113707247B CN 113707247 B CN113707247 B CN 113707247B CN 202111012695 A CN202111012695 A CN 202111012695A CN 113707247 B CN113707247 B CN 113707247B
- Authority
- CN
- China
- Prior art keywords
- data
- treatment
- neural network
- cartilage repair
- model
- 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
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
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/20—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/27—Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G16H15/00—ICT specially adapted for medical reports, e.g. generation or transmission thereof
-
- 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
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- General Physics & Mathematics (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Epidemiology (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Computational Biology (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Evolutionary Biology (AREA)
- Marketing (AREA)
- Pathology (AREA)
- Tourism & Hospitality (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Mathematical Physics (AREA)
- Prostheses (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain, which comprises the following steps: s1, establishing a database based on a block chain; s2, preprocessing original data; s3, processing lost data; s4, model development; s5, generating MSC treatment guidelines for cartilage repair based on machine learning of the blockchain. The invention can accurately predict the treatment result of the patient receiving the MSC cartilage repair treatment according to the condition and the treatment strategy of the patient, thereby providing important reference for a clinician to decide whether and how to perform the MSC treatment on the patient suffering from cartilage damage. In addition, the invention introduces the blockchain technology, can promote the development of traceability and repeatability in the artificial intelligence field, and further relieve a plurality of unrepeatable problems in the currently-popular artificial intelligence field. Therefore, the invention is very suitable for popularization and application.
Description
Technical Field
The invention relates to the technical field of regenerative medicine, in particular to a method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain.
Background
Stem Cell (MSC) therapy is the most promising candidate in regenerative medicine, but inconsistencies in MSC therapeutic efficacy have been reported in various clinical trials. One reasonable reason is the lack of quantitative scientific guidelines for the formulation of personalized MSC treatment strategies. Since the determination of optimal treatment parameters depends on the individual condition of the patient, it is difficult to quantitatively derive a treatment strategy for optimal treatment outcome for the patient by means of conventional biomedical studies alone.
In recent years, in various biomedical fields, the use of machine learning techniques has made significant scientific progress, but few attempts have been made for stem cell therapy, mainly because existing machine learning models, especially neural network models, require a large amount of data support, such as google's Tensorflow, which requires at least more than 1000 lines of data sets to run. Whereas clinical trial data is typically small (a few to tens of data points), if the data set is smaller than the variable number, it cannot be processed. Furthermore, because of the problems with experimental design and data collection, incomplete entries of input information (i.e., lost data) may be included, whereas existing machine learning models typically require a complete data set. Therefore, the existing machine learning technology is utilized to predict the cartilage repair treatment efficacy of the MSC so as to obtain an optimal treatment strategy of personalized quantitative MSC, and the requirements on data, time and cost are very high and almost impossible to realize.
Therefore, there is a need to develop a new machine learning model to predict cartilage repair treatment efficacy of MSCs to achieve the best therapeutic strategy to provide personalized quantitative MSCs for different patients.
Disclosure of Invention
The invention aims to provide a method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of a blockchain, which aims to obtain the optimal rehabilitation effect expected by a specific patient through prediction and a treatment strategy for achieving the aim.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a method for quantitative prediction of cartilage repair rate following mesenchymal stem cell therapy based on machine learning of blockchain, comprising the steps of:
s1, establishing a database based on a block chain: collecting published clinical test reports on MSC cartilage repair therapies, obtaining raw data, and classifying the raw data into data input and data output; simultaneously, storing the report source along with the raw data on a chain;
s2, preprocessing the original data: normalizing the output scores using the different scoring systems to [0,1], wherein 0 represents the most severe injury or pain and 1 represents a completely healthy tissue; meanwhile, according to the existing common sense basis, distributing some initial values to missing data in a database to serve as missing values of preliminary prediction; for example, when the cartilage of a mouse is artificially damaged by using a drill, and the damage depth exceeds the cartilage depth, the cartilage depth damage is calculated as total loss of 100%;
s3, lost data processing: the true value is set to the average of the values present in the dataset, then the following equation is recursively applied until convergence, and finally the result of the convergence is returned:
x (n+1) ＝γx n +(1-γ)f(x n )
wherein x represents a missing value and n represents an iteration step; f (x) n ) Representing predictions about x obtained from a neural network; gamma represents softening parameter and gamma e 0,1]The method comprises the steps of carrying out a first treatment on the surface of the The function f remains fixed in each iteration of the loop;
s4, model development:
(1) Capturing the functional relationship among all variables through an artificial neural network framework;
(2) All x in the database are as close as possible to the function f of the fixed point equation f (x) =x;
(3) Constraining hidden nodes by applying different activation functions;
(4) Treating the state variables, the processing variables and the processing results as inputs and outputs of the neural network, and iteratively improving the estimation of all missing values using a expectation maximization algorithm;
(5) Establishing a neural network model according to the steps (1) - (4);
s5, generating MSC treatment guidelines for cartilage repair based on machine learning of block chain:
(1) Randomly extracting and replacing a group of fixed data sets to enable the group of data to have a more flexible structure to generate more models;
(2) Using the generated model for training the neural network, and simultaneously, recording the model and evaluating the predicted performance; in the training process, the information generated by each iteration of the model is recorded as a hash value to be counted into a chain, the hash value before the next iteration needs to be verified is recorded as a new hash value, and then the new hash value is recorded as a result and stored in the chain;
(3) The overall performance of the predictions is derived from the performance of all the different models, where predictions with greater uncertainty will instruct how to further collect additional data to supplement the database (e.g., stem cells made using cord blood will exhibit greater uncertainty in efficacy than bone marrow-made stem cells under otherwise identical conditions, and more experimental data will be needed in the future to verify the validity of cord blood stem cells); the smaller uncertainty predictions will be used to generate a quantitative guideline for a range of personalized MSC treatments for cartilage repair.
Specifically, in the step S1, the input of data includes a condition variable and a treatment variable; the number of data includes the treatment outcome.
Preferably, in the step S3, γ=0.5.
Further, before proceeding to step S5, the accuracy of the model is evaluated by identifying key descriptors.
Still further, in the step S5, the neural network is iteratively trained by changing a weight matrix of the neural network model.
Compared with the prior art, the invention has the following beneficial effects:
(1) To focus on the most reliable prediction of treatment outcome and to use a relatively small clinical trial database for adequate model training, two unique treatments are provided in the present protocol: the uncertainty of the missing data and the calculation prediction is processed, so that the treatment result of the patient receiving the MSC cartilage repair treatment can be accurately predicted according to the condition of the patient and the treatment strategy. This also optimizes the combination of stem cell therapy and machine learning techniques, providing important references for the clinician to decide whether and how to treat MSC in patients with cartilage damage.
(2) In the invention, when the missing data processing is carried out, the final return is the convergence result, rather than f (x) n ) On the one hand, the convergence result simultaneously contains data internal information and data external machine learning feedback results, which is equivalent to the common prediction by utilizing the correlation in the data and the external neural network learning results; on the other hand, gamma>0 can prevent predicted fluctuations and divergences (γ is preferably 0.5). Therefore, the invention can utilize all information in the database, and provides guarantee for obtaining more reliable modes and improving the prediction quality.
(3) In the invention, after the neural network model is established, the neural network is trained iteratively by changing the weight matrix, which is favorable for minimizing the cost function (also called loss function) and reducing the final prediction cost.
(4) In the prior art, a common neural network can strictly distinguish variables into inputs or inputs, and only one of the variables can be selected, which is not beneficial to further filling of lost data. In the invention, when the model is developed, the state variables, the processing variables and the processing results are regarded as the input and the output of the neural network, and the expectation maximization algorithm is adopted to iteratively improve the estimation of all missing values, so that the most important and accurate association among the missing values is found as far as possible, the accuracy of the original (the state variables, the processing variables as the input and the results as the output) model is further improved, and the missing data is further supplemented in the process. For example, in a common model, state variables and process variables would be split into inputs and the process results would be the outputs (not present in the input training set). The invention takes the processing result as input, such as the processing result, the stem cell number as input, the cartilage damage size as output, and a neural network is trained, wherein the association and prediction can be used for completing the lost data in the cartilage damage.
(5) According to the invention, a group of fixed data sets are randomly extracted and replaced, so that a group of data has a more flexible structure to generate more models, and the neural network is trained. This is also more relevant to the statistical situation and statistically defines the uncertainty of the predicted outcome, thereby ensuring a reliable degree of quantitative guidance for personalized MSC treatment. For example, the present invention is able to produce such guidelines: within 3 months post-surgery, this 60-65 year old patient was expected to receive a dose of 2000 ten thousand bone marrow MSC implants (treatment variable input) with a cartilage area loss of 20-30% (condition variable input) and a maximum repair rate of 80% (output).
(6) The invention combines machine learning and a blockchain technology, and the blockchain technology is mainly used for data acquisition and model generation in the scheme of the invention, so that the reliability and the authenticity of the whole data and model can be effectively improved, and the method specifically comprises the following steps: a. the acquired report and the original data are stored on the chain together, so that the traceability of the data is ensured, and the cost of data forging is greatly increased; b. in the training process of the neural network, information generated by each iteration of the model is recorded as a hash value to be counted into a chain, the hash value before verification is needed in the next iteration, and the result is recorded as a new hash value to be stored in the chain, so that the intermediate algorithm is ensured to be searchable and verifiable without manual intervention, each model after training can find an original data set and training parameters according to the information on the chain, namely, each prediction can track the factor generating the result, and therefore manual modification or fake making is avoided. Thus, the introduction of blockchain techniques in the present invention may promote the development of traceability and repeatability in the field of artificial intelligence, thereby alleviating many of the unrepeatable problems in the currently popular field of artificial intelligence (e.g., researchers do not provide real data and algorithms).
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention.
FIG. 2 is a schematic diagram of possible damage of a patient according to an embodiment of the present invention.
FIG. 3 is a graphical representation of the number of stem cell injections recommended for a patient at median lesion level in an embodiment of the present invention.
Detailed Description
The invention will be further illustrated with reference to the following examples, which are included by way of illustration, but not limitation.
Examples
The invention provides a method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of a blockchain, which aims at predicting the curative effect of MSC transplantation repair cartilage by training a database generated from published clinical test reports, then developing a machine learning model by combining the blockchain and utilizing the interrelation between attributes and the relativity between the attributes. The invention mainly comprises a plurality of processes of establishing a block chain-based database, preprocessing original data, processing lost data, developing a model, evaluating the accuracy of the model and generating an MSC treatment guide for cartilage repair based on machine learning of the block chain.
The above-described flow will be described in detail as shown in fig. 1.
1. Database creation and storage onto blockchain
Published clinical trial reports on MSC cartilage repair therapies (e.g., published reports obtained from PubMed and international cartilage repair association (ICRS) databases, including traumatic cartilage defects, osteoarthritis, etc.) are collected, raw data are obtained and classified as input and output of data. The collected clinical trial reports will be stored along with the raw data on the chain, ensuring the data is searchable.
The input of data includes condition variables such as age, sex, BMI, severity of cartilage damage (diameter and depth of defect before treatment), etc., as well as treatment variables such as cell dose, source, method of delivery, etc. The output of the data will be the treatment outcome, measured by different cartilage repair scores, pain degree scores (e.g., ICRS scoring system, visual analog score of pain (VAS)), etc.
2. Preprocessing of raw data
Output scores using different scoring systems, such as the International Cartilage Repair Society (ICRS) scoring system, the O' droiscoll scoring, the pinada scoring, the Mankin scoring, the international research on Osteoarthritis (OARSI) scoring system, the international commission on knee literature (IKDC) scoring, the pain Visual Analog Scoring (VAS), the knee injury and osteoarthritis outcome scoring (KOOS), the university of cyrthritis index (WOMAC) and the Lyscholm scoring system of cyathromycin and majoram, were normalized to [0,1] for cross-comparison, where 0 represents the most severe injury or pain and 1 represents a fully healthy tissue. Meanwhile, some initial values are assigned to missing data in the database as missing values of preliminary predictions on the basis of existing common knowledge, and these initial values will be used as starting points for iterative machine learning later. For example, not all reports, particularly for Osteoarthritis (OA) trials, will represent cartilage damage in terms of area and depth defect diameter, which will result in "loss" of input data, and then make preliminary guesses for these values based on underlying assumptions (defect size is linearly proportional to OA level).
3. Lost data handling
It is noted above that due to experimental design or data collection problems, the database may contain entries with incomplete input information, and the likelihood of such missing data is greater if the results of studies with acceptable differences in design and purpose are assembled to form the database. For example, in the database of the present invention, osteochondral defect studies have the defect area as common key data for assessing the severity of the injury [14,17]. However, this information is not always provided in osteoarthritis studies [25,48] due to the difficulty in accurately measuring defect areas with complex geometries [25,48], which can lead to a lack of data in the entry.
The design thought of the invention is that potential relativity can exist among different attributes, and missing information can be filled through neural network extraction; a typical neural network requires that each attribute be an input or output of the network, and that all inputs must be provided to calculate a valid output; in contrast, in the present invention, the neural network takes as input known treatment conditions and treatment results (if known), and then outputs predictions of the unknown treatment conditions and treatment results; finally, a neural network loop is applied iteratively to predict unknown treatment conditions and treatment outcomes.
Thus, for any unknown attribute, the missing value is first set to the average of the values present in the dataset. By estimating all values of the neural network, then recursively applying the following equations until convergence, and finally returning the result of convergence:
x (n+1) ＝γx n +(1-γ)f(x n )
wherein x represents a missing value and n represents an iteration step; f (x) n ) Representing predictions about x obtained from a neural network; gamma represents softening parameter and gamma e 0,1]Typically set to 0.5; the function f remains fixed in each iteration of the loop.
4. Model development
After preliminary prediction and processing of the lost data, according to the design thought, firstly, capturing the functional relation among all variables through an artificial neural network framework, then, enabling all x in a database to be as close to a function f of a fixed-point equation f (x) =x as possible, and restraining hidden nodes by applying different activation functions.
The state variables, process variables and process results are then treated as inputs and outputs to the neural network and an expectation maximization algorithm is employed to iteratively refine the estimates of all missing values.
And finally, establishing a neural network model according to the processing.
5. Evaluating accuracy of a model
The invention evaluates the accuracy of the model by identifying key descriptors, and specifically comprises the following steps: principal Component Analysis (PCA) will be performed to identify linear combinations of key descriptors that can use a minimum number of input parameters to grant model accuracy.
Random cross-validation on each training dataset will then be used to select the super-parameters, including the number of hidden nodes per output, the iteration period and the number of training (without reference to the corresponding test dataset).
Accordingly, the performance of the neural network model will be based on the independent validation data and compared to other machine learning methods, such as Random Forest (RF), ensemble matrix factorization (CMF), and multi-objective Deep Neural Network (DNN) models, to further evaluate the accuracy of the neural network model.
6. Blockchain-based machine learning generation of MSC treatment guidelines for cartilage repair
The invention uses machine learning to calculate the uncertainty of the predictions while taking into account the uncertainty of the experiment and the uncertainty of the model itself.
The implementation steps are as follows:
(1) Randomly extracting and replacing a group of fixed data sets to enable the group of data to have a more flexible structure to generate more models;
(2) Using the generated model for training the neural network, and simultaneously, recording the model and evaluating the predicted performance; in this step, the robustness of the activation function, trained weight matrix and superparameter will be tested and benchmarked, and the neural network is iteratively trained by changing its weight matrix to minimize the cost function; meanwhile, the information generated by each iteration of the model is recorded as a hash value to be counted into a chain, the hash value before the next iteration needs to be verified is recorded as a new hash value, and the new hash value is stored in the chain;
(3) Obtaining a predicted overall performance from the performance of all different models, wherein predictions with greater uncertainty will instruct how to supplement the database with additional data of a particular input type; the smaller uncertainty predictions will be used to generate a quantitative guideline for a range of personalized MSC treatments for cartilage repair.
The quantitative guideline content for a series of personalized MSC treatments for cartilage repair includes: the expected optimal therapeutic outcome for a particular patient, the therapeutic strategy to achieve the optimal outcome, the optimal healing after treatment, and the point in time at which minimal pain occurs.
For example, the degree of joint damage (the percentage of damage depth, see x-axis in fig. 2, and the percentage of damage area, see y-axis in fig. 2) is known for patients. The model will optimize the optimal treatment regimen (including stem cell type, number of stem cell injections, etc.) in the background and give the expected results after treatment (see z-axis in fig. 2). The efficacy of the treatment is normalized to the interval 0,1 indicating that the best state of recovery known to date is achieved, 0 indicating no effect. It can be seen that in deep injury areas (e.g., greater than 60% of the injury depth and greater than 80% of the injury area), stem cells do not provide effective recovery to the patient, in which case the patient should consider selecting alternatives other than stem cells, saving time and money costs. Fig. 2 contains all possible damage scenarios.
For a single patient, the lesion situation is known and determined, and the model is optimized according to patient-specific information. FIG. 3 is a graph of the recommended number of stem cell injections (FIG. 3 corresponds to a point in the 3D contour plot) for a patient at the median lesion, and it can be seen that the prognosis works best when injecting twenty-seven million (27 million) stem cells. Recommended doses for general patients are in a set of clinical control experiments for different injected cell numbers (25 mil,50mil,75mil,150 mil) [ reference: gupta PK, chullikana A, rengasamy M, shetty N, pandeY V, agarwal V, et al. effect and safety of adult human bone marrow-determined, cultured, deposited, allogeneic mesenchymal stromal cells (Stempeucel): preclinical and clinical trial in osteoarthritis of the knee joint. Arthritis research & treatment.2016; 18 (1): 301.) in which the 25mil effect of the injection was found to be greater than in the other groups, specific patient information in the experimental group was not available due to experimental ethical requirements. The median lesions used are representative and the trend of decreasing efficacy at higher doses of the model is also consistent with the experimental results. It is worth noting that the model data does not contain any such experimental data and can be considered as a test set in machine learning.
In addition, by analyzing the severity of cartilage damage and in the optimal cartilage repair regimen for a particular patient, the present invention can also determine damage thresholds beyond which the prediction of the present invention will be that the MSC will fail to heal after treatment, taking into account other treatment strategies to avoid unnecessary time and cost losses.
It can be seen that the present protocol is an open-ended study of MSC treatment efficacy based on machine-learned patient and treatment specific information, the success of which will establish quantitative guidelines for clinicians to provide a reference for future strategies for personalized cell therapy. In addition, the invention introduces the blockchain technology, on one hand, the reliability and the authenticity of the whole data and the model can be improved; on the other hand, the development of traceability and repeatability of the artificial intelligence field can be promoted, and therefore the problem that many artificial intelligence fields are not repeatable in the current hot field is relieved.
The above embodiment is only one of the preferred embodiments of the present invention, and should not be used to limit the scope of the present invention, and all the modifications or color changes that are not significant in the spirit and scope of the main body design of the present invention are still consistent with the present invention.
Claims (2)
1. A method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain, comprising the following steps:
s1, establishing a database based on a block chain: collecting published clinical trial reports on MSC cartilage repair therapies, obtaining raw data and classifying the raw data into inputs and outputs of data, the inputs of data comprising condition variables and treatment variables, the outputs of data comprising treatment results; simultaneously, storing the report source along with the raw data on a chain;
s2, preprocessing the original data: normalizing the output scores using the different scoring systems to [0,1], wherein 0 represents the most severe injury or pain and 1 represents a completely healthy tissue; meanwhile, according to the existing common sense basis, distributing some initial values to missing data in a database to serve as missing values of preliminary prediction;
s3, lost data processing: the missing values are set as the average of the values present in the dataset, then the following equations are recursively applied until convergence, and finally the result of the convergence is returned:
x (n+1) ＝γx n +(1-γ)f(x n )
wherein x represents a missing value and n represents an iteration step; f (x) n ) Representing predictions about x obtained from a neural network; gamma represents softening parameter and gamma e 0,1]The method comprises the steps of carrying out a first treatment on the surface of the The function f remains fixed in each iteration of the loop;
s4, model development:
(1) Capturing the functional relationship among all variables through an artificial neural network framework;
(2) All x in the database are as close as possible to the function f of the fixed point equation f (x) =x;
(3) Constraining hidden nodes by applying different activation functions;
(4) Treating the state variables, the processing variables and the processing results as inputs and outputs of the neural network, and iteratively improving the estimation of all missing values using a expectation maximization algorithm;
(5) Establishing a neural network model according to the steps (1) - (4);
then evaluating the accuracy of the model by identifying key descriptors;
s5, generating MSC treatment guidelines for cartilage repair based on machine learning of block chain:
(1) Randomly extracting and replacing a group of fixed data sets to enable the group of data to have a more flexible structure to generate more models;
(2) Using the generated model for training the neural network, and simultaneously, recording the model and evaluating the predicted performance; in the training process, the information generated by each iteration of the model is recorded as a hash value to be counted into a chain, the hash value before the next iteration needs to be verified is recorded as a new hash value, and then the new hash value is recorded as a result and stored in the chain; wherein the neural network is iteratively trained by changing a neural network model weight matrix;
(3) Obtaining predicted overall performance from the performance of all the different models, wherein predictions with greater uncertainty will guide how additional data is further collected to be supplemented into the database; the prediction of less uncertainty will be used to generate a quantitative guideline for a range of personalized MSC treatments for cartilage repair.
2. The method of claim 1, wherein in step S3, γ=0.5.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010909406.7A CN112053750A (en) | 2020-09-02 | 2020-09-02 | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of block chain |
| CN2020109094067 | 2020-09-02 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113707247A CN113707247A (en) | 2021-11-26 |
| CN113707247B true CN113707247B (en) | 2023-09-22 |
Family
ID=73607292
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010909406.7A Pending CN112053750A (en) | 2020-09-02 | 2020-09-02 | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of block chain |
| CN202111012695.1A Active CN113707247B (en) | 2020-09-02 | 2021-08-31 | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010909406.7A Pending CN112053750A (en) | 2020-09-02 | 2020-09-02 | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of block chain |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN112053750A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB201608086D0 (en) * | 2016-05-09 | 2016-06-22 | Sistemic Scotland Ltd | Functional prediction of MSCs |
| CN105903077A (en) * | 2016-04-14 | 2016-08-31 | 山东景源生物科技有限公司 | Preparation method of biological matrix gel material for promoting regeneration and repair of cartilage, gel performance determination method and application |
| CN109381487A (en) * | 2018-12-10 | 2019-02-26 | 天津长和生物技术有限公司 | Application of the human umbilical cord mesenchymal stem cells in preparation prevention and/or treatment osteoarthritis drugs |
| CN109875522A (en) * | 2019-04-22 | 2019-06-14 | 上海健康医学院 | A method of prediction prostate biopsy and root value criterion pathological score consistency |
-
2020
- 2020-09-02 CN CN202010909406.7A patent/CN112053750A/en active Pending
-
2021
- 2021-08-31 CN CN202111012695.1A patent/CN113707247B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105903077A (en) * | 2016-04-14 | 2016-08-31 | 山东景源生物科技有限公司 | Preparation method of biological matrix gel material for promoting regeneration and repair of cartilage, gel performance determination method and application |
| GB201608086D0 (en) * | 2016-05-09 | 2016-06-22 | Sistemic Scotland Ltd | Functional prediction of MSCs |
| CN109381487A (en) * | 2018-12-10 | 2019-02-26 | 天津长和生物技术有限公司 | Application of the human umbilical cord mesenchymal stem cells in preparation prevention and/or treatment osteoarthritis drugs |
| CN109875522A (en) * | 2019-04-22 | 2019-06-14 | 上海健康医学院 | A method of prediction prostate biopsy and root value criterion pathological score consistency |
Non-Patent Citations (4)
| Title |
|---|
| "Artificial Intelligence and Machine Learning in Radiology: Opportunities, Challenges, Pitfalls, and Criteria for Success";James H. Thrall, Xiang Li, Quanzheng Li;《Journal of the American College of Radiology》;第第15卷卷;全文 * |
| 不同间充质干细胞来源的外泌体治疗骨关节炎的研究进展;张季永;崔维顶;;南京医科大学学报(自然科学版)(第05期);全文 * |
| 间充质干细胞源性外泌体在骨科疾病治疗中的作用与应用前景;李嘉;王智慧;吴迪;赵杨;;中国组织工程研究(25);全文 * |
| 间充质干细胞移植治疗膝关节骨关节炎疗效的系统评价与Meta分析;谢保城;吴佳欢;吴巧玲;施伟梅;徐道华;吴龙火;;赣南医学院学报(09);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112053750A (en) | 2020-12-08 |
| CN113707247A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20160253473A1 (en) | Integrated virtual patient framework | |
| CN103116707A (en) | Heart disease intelligent diagnostic method based on case reasoning | |
| CN110841205A (en) | Accurate dose verification method, device and equipment for tumor patient | |
| WO2021151327A1 (en) | Triage data processing method and apparatus, and device and medium | |
| CN116910172A (en) | Follow-up table generation method and system based on artificial intelligence | |
| CN113628761B (en) | Method for predicting treatment efficacy of anti-angiogenesis drug combined with immune checkpoint inhibitor | |
| CN110021399A (en) | A kind of automatic design method of radiotherapy treatment planning | |
| Abdillah et al. | An expert system for diagnosis of rheumatic disease types using forward chaining inference and certainty factor method | |
| CN113707247B (en) | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy based on machine learning of blockchain | |
| CN117711552A (en) | Radiotherapy clinical auxiliary decision-making method and system based on artificial intelligence knowledge base | |
| CN106570325A (en) | Partial-least-squares-based abnormal detection method of mammary gland cell | |
| CN113674820B (en) | Method for quantitatively predicting cartilage repair rate after mesenchymal stem cell therapy by machine learning | |
| CN107050668B (en) | Radiotherapy effect prediction method based on cloud processing | |
| Lambri et al. | Multicentric evaluation of a machine learning model to streamline the radiotherapy patient specific quality assurance process | |
| CN104091094A (en) | Missing value processing method based on multiple fill method and guided by Markov Chain Monte Carlo (MCMC) simulation under Meta analysis numerical value missing state | |
| US12020823B2 (en) | Integrated virtual patient framework | |
| CN115100155A (en) | Method and system for establishing radiation pneumonitis prediction model | |
| Xie et al. | Thyroid disease diagnosis based on feature interpolation and dynamic weighting ensemble model | |
| CN113870984A (en) | AI diagnosis and treatment follow-up system-based standardized diagnosis and treatment center authentication and assessment system | |
| Leonenko et al. | A Modeling Framework for Decision Support in Periprosthetic Joint Infection Treatment | |
| Ghous et al. | Early Detection of Breast Cancer Tumors Using Linear Discriminant Analysis Feature Selection with Different Classification Methods | |
| Sadykova | Web Technologies for Support for Physicians’ Decision Making | |
| Shen et al. | A quantitative quality control method of big data in cancer patients using artificial neural network | |
| Ding et al. | Research on the application of artificial intelligence in the field of depression | |
| Suresh et al. | Prior Cardiovascular Disease Detection using Machine Learning Algorithms in Fog Computing |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20220208 Address after: 200082 room 308-02, building 1, No. 135, Yixian Road, Yangpu District, Shanghai Applicant after: Shanghai Shenyu Racing Technology Co.,Ltd. Address before: No.99, west section of Hupan Road, Xinglong Street, Tianfu New District, Chengdu, Sichuan 610000 China (Sichuan) pilot Free Trade Zone Applicant before: Chengdu Yushi Information Technology Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |