CN112992364A - Cognition-based universal first schizophrenia diagnosis model for teenagers and adults, construction method and diagnosis system - Google Patents

Abstract

The invention provides a cognitive-based diagnosis model, a construction method and a diagnosis system for first schizophrenia, which are universal for teenagers and adults, and relates to the technical field of schizophrenia diagnosis. The diagnosis model is constructed by the following method: identifying redundant samples by using a combined sampling mode integrating an upsampling method and a downsampling method according to MCCB (cognitive test) cognitive test indexes, modeling by using a support vector machine classification algorithm, performing cross validation by using a leave-one-out method, and adjusting model parameters to construct the MCCB cognitive test model; the MCCB cognitive test index does not include a social cognitive index. The diagnosis model of the invention can be universally used for objective diagnosis of adult and juvenile schizophrenia patients, and the repeatability and accuracy of the test result are high.

Description

Cognition-based universal first schizophrenia diagnosis model for teenagers and adults, construction method and diagnosis system

Technical Field

The invention relates to the technical field of schizophrenia diagnosis, in particular to a cognitive-based diagnosis model, a construction method and a diagnosis system for first-onset schizophrenia, which are universal for teenagers and adults.

Background

Schizophrenia is a serious mental disorder, and is clinically manifested by hallucinations, delusions, thought, disorganized speech, and negative symptoms. The global incidence of schizophrenia is about 1%. Clinical studies show that the disease mostly occurs in adolescents and early adulthood, the earlier treatment is the better, and the correct diagnosis in early stage is a key factor for determining whether the schizophrenia can obtain the maximum functional rehabilitation. At present, the diagnosis of the disease is mainly based on symptom manifestations lacking objective basis, however, the clinical manifestations of the disease are various, especially for patients with schizophrenia with juvenile onset, the clinical manifestations are more complicated and variable, and the early diagnosis is difficult.

Currently, no objective diagnostic tool is available clinically for schizophrenia in adolescents and adults, and no ready-made method for reference exists. The existing adult schizophrenia diagnosis model based on cognition has a non-strict modeling process, or is not cross-verified, or is not further verified by a strict test set, or even has no test data set, so that the generalization of the model cannot be well ensured, and the model performance is difficult to reproduce when the model is really applied to clinic. Therefore, there is an urgent clinical need in the psychiatric department to develop an early schizophrenia identification or diagnosis tool based on objective markers, with high accuracy, and applicable to both adolescents and adults.

Disclosure of Invention

Therefore, in order to solve the above problems, there is a need to provide a method for constructing a universal diagnosis model for first-onset schizophrenia of teenagers and adults based on cognition, wherein the constructed diagnosis model can be universally used for objective diagnosis of patients with adult and juvenile schizophrenia, and the test result has high repeatability and accuracy.

A construction method of a universal diagnosis model of first schizophrenia based on cognition for teenagers and adults comprises the following steps:

sampling: collecting a sample, wherein the sample comprises a patient and a healthy person; the patient met the diagnostic criteria of schizophrenia in the American diagnostic and statistical manual for mental disorders, version 5 (DSM-5), and was the first onset; the healthy subject is not in compliance with DSM-5 mental disorder diagnosis criteria either currently or previously;

dividing: dividing a training set from the sample, wherein the training set comprises a patient and a healthy person;

data preprocessing: scoring the testees of the training set by adopting an international general Cognitive testing tool MCCB (MATRICS Consensus Cognitive testing) Cognitive test, wherein the MCCB Cognitive testing indexes do not comprise social Cognitive indexes;

constructing a model: the method comprises the steps of matching the number of samples of patients and healthy persons in a training set by a combined sampling method integrating an up-sampling method and a down-sampling method, modeling by adopting a support vector machine classification algorithm, and adjusting model parameters by leave-one-out cross validation to obtain a universal first-onset schizophrenia diagnosis model for teenagers and adults based on cognition.

Research shows that adults and teenagers with schizophrenia have wide cognitive function damage in early disease stage, the cognitive defect has stable inheritance and is closely related to the functional state of the patients, the cognitive function can be used as the objective characteristic of early recognition of adult and teenagers with schizophrenia, the cognitive function is integrated with machine learning technology, an individualized diagnosis model is expected to be established, and the adult and teenagers with schizophrenia can be effectively recognized in early stage. The neuro-cognitive diagnosis marker is simple and convenient to operate, is not limited by objective conditions such as instrument facilities and the like, is easy to popularize to the basic level, and has relatively stable and reliable detection results.

The invention aims at the current situation that an objective schizophrenia diagnosis model which is generally used for adults and teenagers does not exist in clinic, and establishes a universal schizophrenia diagnosis model for adults and teenagers by a normative data preprocessing and machine learning classification model construction process according to MCCB indexes. According to the invention, social cognition dimensionality is removed on the basis of MCCB indexes, the inventor finds that younger patients are difficult to complete a social cognition module in the MCCB through multipoint data acquisition, and the past adult schizophrenia model based on cognition is often brought into the social cognition dimensionality indexes, so that the younger patients are difficult to further apply to adolescents. Therefore, the social cognition dimension variable is removed in the model construction process, and the feasibility of the model in the actual diagnosis application of the juvenile patient is improved.

Moreover, the invention adopts a joint sampling method, namely an up-sampling method and a down-sampling method, to solve the problem of unbalanced sample size of the training set, which is a possible research confounding factor often encountered in practical research. The up-sampling method can enable a few samples to be amplified to be as many as most samples, but can cause sample noise to be increased, and then the down-sampling method is used for removing data noise generated by the up-sampling method, so that redundant samples influencing identification are effectively reduced, and the classification accuracy of the algorithm is improved.

According to the construction method, social cognitive indexes which are not suitable for younger teenagers are removed, and the feasibility of the model in the practical diagnosis application of teenager patients is improved. Generally, the number of samples of patients and healthy persons (i.e. healthy controls) in a training set is different, i.e. the samples are unbalanced, and the problem of unbalanced samples may cause that when a model is trained, the model tends to select most classes in data as a classification basis, so that the trained model loses application value; therefore, the model is constructed by adopting a joint sampling method, the few samples are expanded to be as many as the majority samples by utilizing an up-sampling method, and the data noise generated by the up-sampling method is removed by utilizing a down-sampling method, so that redundant samples influencing discrimination are effectively reduced, and the classification accuracy of the algorithm is improved. The diagnosis model constructed by the method can be universally used for objective diagnosis of adult and juvenile schizophrenia patients, the repeatability of the test result is high, the classification accuracy of the patients and the contrast in the test set of adults and minors is more than 84%, and the generalization capability of the diagnosis model is strong.

"Diagnostic criteria for schizophrenia" refers to the Diagnostic criteria for schizophrenia in the United states handbook of Mental Disorders (5 th Edition), diagnosed by 2 psychiatric physicians and physicians of above title.

In one embodiment, the training set includes adult patients and adult healthy individuals.

In one embodiment, the training set includes adult patients, underage patients, adult healthy individuals, and underage healthy individuals.

In one embodiment, the model building step is followed by a verification step: the samples except the training set in the dividing step are used as a test set, the diagnosis model of the first schizophrenia is applied to the test set, and the classification effect of the patients and the healthy persons is verified; the test set includes an adult test set and a minor test set.

Preferably, the number of test sets is at least two. The two test sets are adopted to test the model, so that the universality of the diagnosis model can be better reflected.

In one embodiment, the sample collection sites of the training set and the test set are different.

In one embodiment, the number of the sample sources in the training set is at least 2, and the training set adopts samples from multiple sources, so that the model can be trained sufficiently, and the accuracy is improved.

The sample sources refer to areas where samples are collected or hospitals, the samples from different sources are adopted for training and testing, the data independence is good, the constructed model is not limited by areas, the accuracy of cross-regional diagnosis results is high, and the method is favorable for popularization and use in different areas.

The training set and the test set adopted by the invention have good data independence, the training set and the test set are respectively from different centers in the country, and the cross validation and joint sampling process in the training process is only applied to the training set, so that the data independence of the training set and the test set is strictly ensured.

In one embodiment, the MCCB cognitive test metrics include: speed of information processing, attention alertness, working memory, word learning, visual learning, reasoning, and problem solving.

In one embodiment, the MCCB cognitive test metrics include: symbol coding, animal naming classification fluency, connection testing A, attention alertness, working memory, word learning, visual learning, reasoning and problem solving.

By adopting the MCCB cognitive test indexes and combining the model construction means, the accuracy of the obtained diagnosis model in diagnosing the first schizophrenia of adults and minors is higher.

In one embodiment, the upsampling method is a SMOTE oversampling method.

In one embodiment, the down-sampling method is a Tomek Link method.

The invention also provides a cognitive-based diagnosis model for the first schizophrenia, which is universal for teenagers and adults and is obtained by adopting the construction method.

The diagnosis model of the invention can be universally used for objective diagnosis of adult and juvenile schizophrenia patients, the repeatability of the test result is high, and the classification accuracy of the patients and the contrast in the test set of adults and minors is more than 84 percent, which shows that the diagnosis model has strong generalization capability.

The present invention also provides a diagnostic system for common use with first-onset schizophrenia, the diagnostic system comprising:

the data acquisition module is used for acquiring MCCB cognitive test index test scores of the testee;

the analysis module inputs the index scores of the testees into the diagnosis model and analyzes the index scores;

and the output module is used for outputting the diagnosis result of the analysis module.

The diagnosis system adopts the diagnosis model of the invention, can efficiently carry out individual-based objective diagnosis on a testee, and improves the diagnosis efficiency and the stability of the diagnosis result.

Compared with the prior art, the invention has the following beneficial effects:

the diagnosis model solves the current situation that no objective diagnosis model is universally used for adults and teenager schizophrenia patients in clinic; the test result has high repeatability, and the classification accuracy of the model on patients and healthy controls in the test sets of adults and minors is more than 84% after the test of the test sets, so that the model has strong generalization capability and high result repeatability.

The construction method of the invention eliminates social cognition index which is not suitable for younger teenagers, and improves the feasibility of the model in the practical diagnosis application of teenager patients; and moreover, by adopting a joint sampling method, firstly, the few samples in the training set are amplified to be as many as the many samples in the training set by utilizing an up-sampling method, and then, data noise generated by the up-sampling method in the training set is removed by utilizing a down-sampling method, so that the problem of sample imbalance is effectively solved, redundant samples influencing discrimination in the training set are reduced, and the classification accuracy of the algorithm is improved.

Drawings

FIG. 1 is a flow chart of construction of a diagnostic model in an embodiment;

FIG. 2 is a diagram of a test confusion matrix for a test set of minors;

wherein SP is a standardized patient and HC is a healthy control; the horizontal SP and HC are real classification results of the original data; vertically arranging SP and HC as diagnosis classification results;

FIG. 3 is a diagram of a test confusion matrix for an adult test set;

wherein SP is a standardized patient and HC is a healthy control; the horizontal SP and HC are real classification results of the original data; the vertical SP and HC are diagnosis and classification results.

Detailed Description

To facilitate an understanding of the invention, a more complete description of the invention will be given below in terms of preferred embodiments. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

Example 1

Firstly, collecting a research object.

The model samples are from multiple centers in China, and are respectively responsible for recruiting health controls and first schizophrenia patients and health controls treated by the department of academy and clinic from the following four units: shanghai mental health center (54 patients and 48 healthy controls), department of encephalic hospital of Guangzhou medical university (101 patients and 89 healthy controls), first department of Zhengzhou university (107 patients and 53 healthy controls), Beijing Return to Longguan Hospital (71 patients).

Patients with first-onset schizophrenia entered the group criteria:

1) in compliance with the Diagnostic criteria of schizophrenia in the Diagnostic and Statistical Manual of Mental Disorders (5 th Edition), 2 psychologists and the above-named doctors made a consistent diagnosis;

2) age 13-45 years;

3) the first attack is not treated by psychiatric drugs or the cumulative exposure of psychiatric drugs (including antipsychotic drugs, antidepressant drugs, mood stabilizer, etc.) for more than 1 year.

Exclusion criteria for first-onset schizophrenic patients:

1) pregnant or lactating women;

2) combining major somatic diseases affecting cognitive performance and mental development retardation;

3) organic mental disorders and neurological diseases;

4) drug or alcohol abuse/dependence;

5) schizoaffective disorder, depression or bipolar disorder.

Healthy control group entry criteria:

1) mental health, current and past no mental disorder that meets the DSM-5 diagnostic criteria;

2)13-45 years old;

3) sign an informed consent.

Healthy control exclusion criteria:

1) the two generations have family history of mental disorder;

2) pregnancy and lactation;

3) combining major somatic and neurological diseases that affect cognitive performance.

The study was approved by the ethical committee of the brain hospital affiliated to Guangzhou medical university, and the subjects or their guardians signed informed consent.

And II, cognitive assessment.

Currently, the most commonly used in the assessment of Cognitive function in schizophrenia is the assessment to improve cognition in schizophrenia and the Cognitive test suite (MCCB) of therapeutic studies. The test is evaluated using the MCCB scale, which includes 7 cognitive domain scores: information processing speed, attention, alertness, working memory, word learning, visual learning, reasoning and problem solving, social cognition.

In this embodiment, optimized MCCB cognitive test indexes are adopted: symbol coding, animal naming classification fluency, connection testing A, attention alertness, working memory, word learning, visual learning, reasoning and problem solving.

And thirdly, constructing a model.

1. Data pre-processing

Index extraction: as teenagers cannot cooperate with and finish the social cognition module when the reading and understanding ability of the teenagers is limited by more than one third, three modules (symbol coding, animal naming classification fluency, connection test A), attention alertness, working memory, word learning, visual learning, reasoning and T-score (T-score) of problem solving are extracted at the information processing speed after correcting the influence factors such as age, gender, education age and the like, and 8 indexes are used as indexes of the incorporated model, and the indexes are evaluated according to a conventional MCCB cognitive test method.

2. Model construction method

1) Division of training and test sets: a total of 232 patients and 101 healthy controls of Shanghai mental health center, first subsidiary hospital of Zhengzhou university, Beijing-huilongguan hospital were used as a training set, 56 adult patients and 61 healthy controls of Guangzhou medical university subsidiary brain hospital were used as an adult test set, and 45 minors and 28 minors of Guangzhou medical university subsidiary brain hospital were used as minor test sets.

2) The algorithm and the construction process are as follows: the method comprises the steps of enabling the sample amount of a patient in a training set and a sample amount of a health control sample to be unbalanced, enabling the sample amount of the patient in the training set and the sample amount of the health control sample to be matched through a joint sampling method, namely, firstly adopting an SMOTE oversampling method, then effectively reducing redundant samples influencing identification through a Tomek Link method, finally adopting a support vector machine classification algorithm for modeling, and adopting a leave-one-out method for cross validation and adjustment of model parameters.

3) And (3) verification: and respectively applying the model to an adult test set and a minor test set to verify the classification effect of the patient and the health control.

And fourthly, model diagnosis results.

The model construction flow of this embodiment is shown in fig. 1.

And inputting the test result of the cognitive test index into the model for calculation to obtain a diagnosis result. In this embodiment, the calculation process and the result output are completed through the svc.precision _ function in the scimit-spare svm algorithm encapsulation packet, and it can be understood that other commercial or open source statistics tools with similar functions in the conventional technology may also be used for implementation.

The test results (confusion matrix) for the minor test set are shown in fig. 2, and the test results (confusion matrix) for the adult test set are shown in fig. 3.

In the juvenile test set, the accuracy of the model to distinguish patients from healthy controls was 84.93%, the sensitivity was 88.89%, the specificity was 78.57%, and the accuracy of the model to identify patients was 86.96%.

In the adult test set, the accuracy of the model to distinguish patients from healthy controls was 84.62%, the sensitivity was 69.64%, the specificity was 98.36%, and the accuracy of the model to identify patients was 97.50%.

Example 2

A construction method of a diagnosis model is basically the same as the construction steps of the model in the embodiment 1, and is different in that in the index extraction step, the model is constructed by adopting 6 cognitive dimension indexes including information processing speed, attention alertness, working memory, word learning, visual learning, reasoning and problem solving.

The information processing speed is a comprehensive score of symbol coding, animal naming classification fluency and connection test A, and the comprehensive score is a general index which can only provide one-dimensional information. In example 1, symbol encoding, animal naming classification fluency and connection test A are three independent indexes.

And (3) testing results: adult test concentration, the accuracy of the model to distinguish patients from healthy controls was 82.05%; in the juvenile test set, the accuracy of the model to distinguish patients from healthy controls was 82.19%.

Comparative example 1

A diagnostic model construction method is basically the same as the model construction steps in embodiment 1, except that in the model construction method steps, modeling is performed by using a support vector machine algorithm without a joint sampling process.

And (3) testing results: in the adult test set, the best accuracy of the model to distinguish patients from healthy controls was 70.94%; in the juvenile test set, the best accuracy for the model to distinguish patients from healthy controls was 69.86%.

Comparative example 2

A method for constructing a diagnostic model, which has substantially the same steps as those of the model construction in example 1, except that a classical machine learning linear model-logistic regression classification algorithm is used for modeling.

And (3) testing results: adult test concentration, the accuracy of the model to distinguish patients from healthy controls was 71.79%; in the juvenile test set, the accuracy of the model to distinguish patients from healthy controls was 73.97%.

Comparative example 3

A construction method of a diagnosis model is basically the same as the model construction steps in the embodiment 1, and is different in that a classical machine learning nonlinear model-random forest classification algorithm is adopted for modeling.

And (3) testing results: adult test concentration, the accuracy of the model to distinguish patients from healthy controls was 69.23%; in the juvenile test set, the accuracy of the model to distinguish patients from healthy controls was 69.86%.

The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A construction method of a universal diagnosis model of first schizophrenia based on cognition for teenagers and adults is characterized by comprising the following steps:

sampling: collecting a sample, wherein the sample comprises a patient and a healthy person; the patients are in accordance with the diagnosis standard of schizophrenia of 5 th edition of the American Manual of diagnosis and statistics of mental disorders, and are in the first onset; the healthy subject is not in compliance with DSM-5 mental disorder diagnosis criteria either currently or previously;

dividing: dividing a training set from the sample, wherein the training set comprises a patient and a healthy person;

data preprocessing: the method comprises the steps of adopting an international universal cognitive testing tool MCCB (cognitive test kit) for cognitive testing of schizophrenia to score testees in a training set, wherein indexes of the MCCB cognitive testing do not comprise social cognitive indexes;

constructing a model: the method comprises the steps of matching the number of samples of patients and healthy persons in a training set by a combined sampling method integrating an up-sampling method and a down-sampling method, modeling by adopting a support vector machine classification algorithm, and adjusting model parameters by leave-one-out cross validation to obtain a universal first-onset schizophrenia diagnosis model for teenagers and adults based on cognition.

2. The construction method according to claim 1, wherein the training set comprises adult patients, adult healthy people.

3. The construction method according to claim 1 or 2, characterized in that the model construction step is followed by a verification step: the samples except the training set in the dividing step are used as a test set, the diagnosis model of the first schizophrenia is applied to the test set, and the classification effect of the patients and the healthy persons is verified; the test set includes an adult test set and a minor test set.

4. The construction method according to claim 1, wherein the sample collection sites of the training set and the test set are different.

5. The method of constructing according to claim 1, wherein the MCCB cognitive test metrics comprise: speed of information processing, attention alertness, working memory, word learning, visual learning, reasoning, and problem solving.

6. The method of constructing according to claim 1, wherein the MCCB cognitive test metrics comprise: symbol coding, animal naming classification fluency, connection testing A, attention alertness, working memory, word learning, visual learning, reasoning and problem solving.

7. The construction method according to any one of claims 1 to 6, wherein the upsampling method is a SMOTE oversampling method.

8. The method of claim 7, wherein the downsampling method is a Tomek Link method.

9. A cognitive-based diagnosis model for first schizophrenia universal for teenagers and adults by using the construction method of any one of claims 1 to 8.

10. A system for diagnosing first-onset schizophrenia, the system comprising:

the data acquisition module is used for acquiring MCCB cognitive test index test scores of the testee;

an analysis module for inputting the subject index score into the diagnostic model of claim 8 or 9 and performing analysis;

and the output module is used for outputting the diagnosis result of the analysis module.

Images