Adult ADHD evaluation model and evaluation system

Abstract

The invention discloses an adult ADHD evaluation model and evaluation system, and belongs to the technical field of machine learning diagnosis models. The construction method of the model comprises the following steps: collecting basic information of a sample, wherein the basic information comprises gender, age, educated age limit and marriage state, economic condition, smoking and drinking history, mental disease family history, mental disease attack age and mental disease attack frequency; the collected information is preprocessed and divided into a training set and a test set; inputting the basic information in the training set and pre-marked data into a CNN (Convolutional Neural Network) model for training; a learning rate attenuation strategy is configured in the CNN convolutional neural network model, namely alpha = alpha 0-kt, alpha 0 is an initial learning rate, k is an attenuation rate, and t is the number of iterations of training; the CNN convolutional neural network model is constructed and trained through TensorFlow; inputting the basic information in the training set and pre-marked data into a CNN model for training; and adjusting weight parameters to obtain an adult ADHD evaluation model.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning diagnostic models, and in particular to an adult ADHD assessment model and an assessment system. Background Art

[0002] Adult ADHD is a neurodevelopmental disorder characterized by age-inappropriate distractibility, overactivity and emotional impulsivity, accompanied by cognitive impairment and learning difficulties, and normal or near-normal intelligence. It is estimated that the prevalence of ADHD in children is 4%-7%, and among this group of people, 15%-65% will continue to suffer from the disease in adulthood. Moreover, adult ADHD is not necessarily a continuation of childhood ADHD, because a considerable number of adults with ADHD have no history of the disorder in childhood. The general population prevalence of adult ADHD is estimated to be 2.5%, and adults with ADHD show the following symptoms: inattention to details, difficulty organizing tasks and activities, excessive talking or fidgeting, difficulty relaxing, overworking, forgetfulness, and distraction. However, despite the relatively high incidence of ADHD in adults, it is often not discovered in patients who seek medical treatment. Adult ADHD is associated with severe psychosocial dysfunction, which can lead to serious personal and social costs and negatively affect the individual's self-esteem and interpersonal relationships. In a community sample survey study of 1001 adults, people with ADHD were significantly more likely to be divorced and were significantly less satisfied with their personal, social, and professional lives. ADHD is associated with learning difficulties, college students with ADHD have lower average grades, some students need to retake grades, and there are higher rates of suspension and dropout. People with adult ADHD face challenges in time management, organization, and self-regulation, which may lead to employment and financial problems. Adult ADHD is associated with poor driving ability, traffic tickets, and high rates of motor vehicle accidents. A recent study found that Japanese adult ADHD patients visited the doctor 10 times more often than non-ADHD controls, and the emergency room and hospitalization rates were 3 times that of the control group. In Denmark, ADHD patients have a lower life expectancy and more than twice the risk of death than adults without ADHD. This is mainly attributed to accidental deaths caused by ADHD-related characteristics such as risky behavior. The harmful effects of adult ADHD on health and safety make it more urgent to properly recognize and manage this disease. The present invention is mainly used for the diagnosis of adult ADHD.

[0003] The current assessment of adult ADHD includes self-rating scales and peer-rating scales. The self-rating scale is the Adult ADHD Symptom Assessment Scale, which is the World Health Organization Adult ADHD Self-Report Scale (ASRS) authorized by Dr. Kessler's team at Harvard Medical School. It includes two parts, A and B. Part A assesses attention deficit, and Part B assesses impulsivity / hyperactivity. The peer-rating scale is an interview assessment, which also includes two parts: attention deficit and impulsivity / hyperactivity. Among them, the self-rating scale has relatively fewer questions to be answered by the assessee, and the assessment is faster, but the accuracy is poor. Although the interview assessment is relatively accurate, the assessee has to answer more questions, the questions are more complicated, and it takes a long time. Overall statistics show that the probability of the assessee being finally confirmed to have adult ADHD is relatively low, if the cost of direct interview assessment is too high. And even if the interview assessment score is too high, a diagnosis and treatment plan cannot be issued, and further quality of life assessment is required. The whole process is time-consuming and cumbersome. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention proposes an adult ADHD assessment model and an assessment system.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The first aspect of the present invention relates to a method for constructing an adult ADHD assessment model, comprising the following steps:

[0007] Basic information of the samples was collected, including gender, age, years of education and marital status, economic status, smoking and drinking history, family history of mental illness, age of onset of mental illness, and number of mental illness episodes;

[0008] Preprocess the collected information and divide it into training set and test set;

[0009] Inputting the basic information in the training set and the pre-labeled data into the CNN convolutional neural network model for training; the CNN convolutional neural network model is configured with a learning rate decay strategy, that is, alpha=alpha0-kt, where alpha0 is the initial learning rate, k is the decay rate, and t is the number of training iterations; the CNN convolutional neural network model is constructed and trained by TensorFlow;

[0010] The weight parameters were adjusted to obtain the adult ADHD assessment model.

[0011] Optionally, the data ratio of the training set to the test set is 8:1.

[0012] Optionally, after collecting the basic information, data denoising is performed on the basic information.

[0013] The second aspect of the present invention relates to an adult ADHD assessment model constructed by the above-mentioned method for constructing an adult ADHD assessment model.

[0014] A third aspect of the present invention relates to an adult ADHD assessment system, comprising:

[0015] Information input module, to obtain basic information of the evaluation object;

[0016] A pre-assessment module, loaded with the adult ADHD assessment model of claim 4, capable of receiving the basic information and used to calculate an assessment score for whether the person suffers from ADHD;

[0017] Interview assessment module, used to assess attention deficit or assess risk score of impulsivity / hyperactivity through interview;

[0018] And, a quality of life assessment module calculates an assessment risk score based on the quality of life of the assessment subject.

[0019] Optionally, a feedback module is further included, which is used to receive the final diagnosis result and feed it back to the adult ADHD assessment model to adjust the parameters of the adult ADHD assessment model.

[0020] Optionally, the method for calculating the risk score for attention deficit or impulsivity / hyperactivity includes: calculating the risk score using the Adult ADHD Diagnostic Interview 2.0 Chinese Revised Version and its standards.

[0021] Optionally, the quality of life assessment module includes assessments of life productivity, mental health, interpersonal relationships and outlook on life.

[0022] Optionally, each item in the quality of life assessment module is scored using a Likert scale.

[0023] Optionally, the automation technology for adult ADHD assessment application includes electronic information data collection human-computer interaction, assessment analysis, report presentation, and graded assessment jump to reduce unnecessary assessments.

[0024] Optionally, the reward for the estimation model intensive training adopts a comprehensive strategy score of accurate evaluation and follow-up visit of the person being evaluated.

[0025] Optionally, the precise assessment includes an interview assessment and a quality of life assessment, wherein the interview assessment includes two parts: attention deficit and impulsivity / hyperactivity, and the quality of life assessment includes four factors A, B, C, and D, which respectively assess life productivity, mental health, interpersonal relationships, and outlook on life. If the depression and anxiety scores in mental health are high, further anxiety and depression scale assessments are performed.

[0026] Beneficial effects of the present invention:

[0027] The present invention adopts a hierarchical assessment strategy, first pre-assessment and then precise assessment. The pre-assessment gathers the basic information of the person being assessed and the self-assessment scale information, and uses a large amount of data to train the neural network model, making the pre-assessment system of the present invention more accurate than the self-assessment scale.

[0028] The system of the present invention continuously performs intensive training on the prediction model during use, so that the prediction system becomes more and more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below in conjunction with the accompanying drawings.

[0030] Figure 1 A schematic diagram of the system evaluation process and pre-evaluation system reinforcement learning process for this application;

[0031] Figure 2 This is a schematic diagram of the construction of the pre-assessment model Nomogram of this application; the nomogram discriminant model of the probability of co-occurrence of adult ADHD in patients with various mental disorders. Among them, A: nomogram for predicting the co-occurrence of generalized anxiety disorder and adult ADHD; B: nomogram for predicting the co-occurrence of MDD and adult ADHD; C: nomogram for predicting the co-occurrence of bipolar disorder and adult ADHD; D: nomogram for predicting the co-occurrence of SCZ and adult ADHD.

[0032] Figure 3 The following are the prediction validation diagrams of the pre-assessment model of this application in four common mental illnesses, and the calibration diagrams of the adult ADHD probability diagrams of patients with various mental disorders. A. Hosmer-Lemeshow goodness-of-fit test results of the nomogram of the GAD group; B. Hosmer-Lemeshow goodness-of-fit test results of the nomogram of the MDD group; C. Hosmer-Lemeshow goodness-of-fit test results of the nomogram of the BD group; D. Hosmer-Lemeshow goodness-of-fit test results of the nomogram of the SCZ group.

[0033] Figure 4 The workflow diagram of the pre-assessment model for this application;

[0034] Figure 5 This is a training flow chart of the pre-evaluation model for this application;

[0035] Figure 6 The selection process of different assessment items for adult ADHD for this application;

[0036] Figure 7 A self-assessment scale for this application;

[0037] Figure 8 A quality of life assessment questionnaire for this application. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Figure 1 It is a schematic diagram of the entire system evaluation process and the pre-evaluation system reinforcement learning process. The evaluation process is as follows:

[0040] First, basic information of the person being evaluated is collected, including gender, age, years of education and marital status, economic status, smoking and drinking history, family history of mental illness, age of onset of mental illness, number of mental illness attacks, and other general personal information and clinical data information. A trial of 3,772 samples collected from 29 centers across the country found that the model containing the above personal information and clinical characteristics can better judge whether adult ADHD exists, especially in common mental illnesses such as generalized anxiety disorder (GAD), major depressive disorder (MDD), bipolar disorder (BD), schizophrenia (SCH) to judge whether there is comorbid adult ADHD ( Figure 2 , Figure 3 ), the above-mentioned diseases are diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5).

[0041] Second, conduct a preliminary assessment on the person being assessed. During the preliminary assessment, the person being assessed needs to complete a simple self-assessment scale, including two items: assessing attention deficit and assessing impulsivity / hyperactivity. Figure 4 The basic information and ASRS self-assessment scale information are combined and inferred using the pre-assessment model to obtain the pre-assessment scores of the two items. The pre-assessment model is obtained by machine learning training using a large amount of accurately labeled data in advance.

[0042] In some specific embodiments of the present invention, the training data of the pre-assessment model is derived from characteristic information collection of relevant patients by multiple hospitals.

[0043] The specific steps of training are: first pre-process the collected data to clean up the data that is obviously inconsistent with common sense; then import a large part of the pre-processed data and the corresponding diagnosis results into the neural network model for training to obtain a pre-evaluation model; then use the remaining feature data as input to import the pre-evaluation model, compare the results obtained with the doctor's diagnosis results of this part of the data, calculate the loss function, and verify the performance of the model. Finally, by adjusting the weight parameters of the model and training repeatedly, the best performance pre-evaluation model is obtained and saved for prediction and reasoning.

[0044] Training model architecture: The open source TesorFlow AI model Python library is used to define the CNN convolutional neural network model. The CNN convolutional neural network model is chosen because it has better parallel processing and generalization capabilities, higher reasoning efficiency during use, and can continuously improve model performance in the subsequent reinforcement learning process.

[0045] Specific steps: Collect a large amount of data from multiple hospitals; Write a program to define the CNN model; Use the collected basic information to train a high-performance model; Save the model and use it for subsequent reasoning. The learning rate adjustment method is used to improve the performance of the model. A linear learning rate decay strategy is adopted, that is, alpha = alpha_0-kt, where alpha_0 is the initial learning rate, k is the decay rate, and t is the number of training iterations. Repeated training makes the model have better training effect and convergence speed.

[0046] 3. If the score of a certain pre-assessment does not exceed the threshold, a relevant assessment report will be issued.

[0047] Fourth, if the score of a pre-assessment exceeds the threshold, the person being assessed will be interviewed and assessed. The interview assessment also includes two items: attention deficit assessment and impulsivity / hyperactivity assessment, using the Diagnostic Interview for Adult ADHD 2.0 Chinese Revised Version (DIVA2.0-CR). However, only one of them may be needed at this stage, because the pre-assessment score of the other item of the person being assessed does not exceed the threshold.

[0048] 5. If the evaluation score of a certain interview does not exceed the threshold, a relevant evaluation report will be issued.

[0049] 6. If the score of a certain interview assessment exceeds the threshold, the person's quality of life will be assessed using the AAQoL tool specifically for adult ADHD quality of life assessment, and an assessment report will be issued. The content of the assessment form is as follows: Figure 7 shown.

[0050] like Figure 8 As shown, the AAQoL is a 29-item questionnaire designed to assess quality of life in adults with ADHD.

[0051] It consists of four subscale scores:

[0052] (1) Life productivity (11 items): including keeping a clean and tidy home, balancing your financial income and expenditure, remembering important things, completing your shopping (such as food, clothes or household items), paying attention to others when interacting with them, juggling multiple things, completing tasks on time, completing your own tasks (work or family), starting tasks that you are not interested in, remembering the location of important items (such as keys, wallet), and requiring too much effort to complete things.

[0053] (2) Mental health (6 items): feeling overwhelmed, anxious, depressed, feeling tired, your mood swings (ups and downs), and your overreactions to difficult or stressful situations.

[0054] (3) Interpersonal relationships (5 items): being annoyed by others, relationships are strained, you cannot meet other people's expectations of you (at home or work), people are disappointed in you, and you do not have enough time to spend with others.

[0055] (4) Outlook on life (7 items): Your energy is used well (with good returns), you enjoy spending time with others, you manage your life successfully, you get what you want, you are a good person, people enjoy being with you, and your intimate relationships are progressing well.

[0056] Each item was rated by the patient using a 5-point Likert scale, ranging from “not / never” (1) to “extremely / very often” (5).

[0057] To derive the overall and subscale scores, item scores were converted to a 0–100 scale.

[0058] The sum of the item scores divided by the item count produced the subscale and overall scores.

[0059] Note: If >1 item is missing in a subscale score, the subscale score is considered missing. If >3 items are missing in the total score, it is considered missing. The higher the score, the higher the quality of life.

[0060] The pre-evaluation system reinforcement learning process is as follows:

[0061] 1. Based on the evaluation report issued during the evaluation process and the comparison with the patient follow-up results, the evaluation results of the pre-evaluation system are scored. If the patient follow-up results are consistent with the pre-evaluation system results, reward points will be given, otherwise penalty points will be given.

[0062] Second, collect the basic information of the person being evaluated, the self-assessment scale information, and the evaluation score of the pre-assessment system, and feed them back to the pre-assessment system for reinforcement learning.

[0063] Figure 5 This is a diagram of the training process of the pre-evaluation model. The steps are as follows:

[0064] 1. Collect a large amount of data from the target population, mainly people suspected of having adult ADHD, collect their basic information, and conduct self-assessment and peer-assessment scales for them.

[0065] 2. Do data preprocessing. This mainly includes data denoising or normalization. Data denoising refers to the data that is obviously inconsistent with the facts during the data collection process, such as data that a person's age is filled in as 1000 years old. Data normalization is for the convenience of training. For example, gender can be represented by 0 for male and 1 for female.

[0066] 3. Classify the data, 80% of the data is used for training and 20% of the data is used for verification.

[0067] 4. Repeated training and verification to obtain a better pre-assessment model.

[0068] Figure 6 This is the selection process for different assessment items for adult ADHD:

[0069] 1. Conduct a preliminary assessment of attention deficit;

[0070] Second, if the attention deficit pre-assessment score does not exceed the threshold, an attention deficit assessment report will be issued;

[0071] Third, if the attention deficit pre-assessment score exceeds the threshold, an attention deficit interview assessment is conducted;

[0072] 4. Conduct an impulsivity / hyperactivity pre-assessment;

[0073] 5. If the impulsivity / hyperactivity pre-assessment score does not exceed the threshold, an impulsivity / hyperactivity assessment report will be issued;

[0074] 6. If the pre-assessment score for impulsivity / hyperactivity exceeds the threshold, an impulsivity / hyperactivity interview assessment is performed;

[0075] 7. If the scores of both interview assessments do not exceed the threshold, an assessment report will be issued;

[0076] 8. If one or two interview assessment scores exceed the threshold, a quality of life assessment will be conducted. If the mental health depression and anxiety scores in the quality of life are high, further depression and anxiety scale assessments will be conducted and a final assessment report will be issued.

[0077] It should be noted that the model construction method of the present invention is only a data aggregation and processing method, and is not practical as a diagnostic method. The model obtained can be used as a software product. The pre-evaluation model of the present invention can be combined with the evaluation steps of this application, or with other existing evaluation indicators to obtain a more accurate evaluation result.

[0078] The evaluation system in the embodiment of the present invention adopts a hierarchical evaluation strategy, first pre-evaluation and then precise evaluation. The pre-evaluation combines the basic information of the person being evaluated and the self-evaluation scale information, and uses a large amount of data to train the neural network model, so that the pre-evaluation system of the present invention is more accurate than the self-evaluation scale.

[0079] In addition, the system continuously performs reinforcement training on the estimation model during use, making the pre-evaluation system more and more accurate. Reinforcement learning is needed because the information factors considered in the pre-evaluation system may change with time and environment, and their impact on adult ADHD may also change. For example, the economic income factor, as the economy in the environment of the person being evaluated changes or people's perception of economic conditions changes, for example, people used to like economic comparisons, but later stopped comparing, then the impact of economic income on adult ADHD has changed. If the pre-evaluation system trained with big data is still used at this time, the evaluation results will be inaccurate. Therefore, reinforcement learning is needed to continuously optimize the parameters of the neural network model.

[0080] In addition, the present invention also uses software to automatically separate different assessment items, that is, the assessment of attention deficit and the assessment of impulsivity / hyperactivity are separated. If the pre-assessment shows that one of the items has a low score, the item will not be accurately assessed. Therefore, the entire assessment system is fast and accurate.

[0081] The entire process method in the above embodiment can be made into a computer-executable program, and the necessary data and the program are stored in a computer-readable storage medium. By running the program, the above steps can be executed and the corresponding technical effects can be achieved.

[0082] It should also be noted that the pre-assessment model trained in the above embodiment is a software product that can realize the preliminary assessment function, and the storage medium storing the pre-assessment model or the computer device capable of running the pre-assessment model are products that can realize the technical solution and technical effects of this application. Therefore, the training process of the pre-assessment model of this application belongs to the method of manufacturing or constructing the software product, and does not belong to the diagnostic method; while the method of using or running the pre-assessment model belongs to the method of actually performing the diagnosis and obtaining the diagnostic results or risks.

[0083] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0084] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for constructing an adult ADHD assessment model, characterized in that: The following steps are involved: Basic information of the samples was collected, including gender, age, years of education and marital status, economic status, smoking and drinking history, family history of mental illness, age of onset of mental illness, and number of mental illness episodes; Preprocess the collected information and divide it into training set and test set; Inputting the basic information in the training set and the pre-labeled data into the CNN convolutional neural network model for training; the CNN convolutional neural network model is configured with a learning rate decay strategy, that is, alpha=alpha0-kt, where alpha0 is the initial learning rate, k is the decay rate, and t is the number of training iterations; the CNN convolutional neural network model is constructed and trained by TensorFlow; The weight parameters were adjusted to obtain the adult ADHD assessment model.

2. The method for constructing an adult ADHD assessment model according to claim 1, characterized in that: The data ratio of the training set to the test set is 8:

1.

3. The method for constructing an adult ADHD assessment model according to claim 1, characterized in that: After the basic information is collected, data denoising is performed on the basic information.

4. An adult ADHD assessment model constructed by the method for constructing an adult ADHD assessment model according to any one of claims 1 to 3.

5. An adult ADHD assessment system, characterized in that: include: Information input module, to obtain basic information of the evaluation object; A pre-assessment module, loaded with the adult ADHD assessment model of claim 4, capable of receiving the basic information and used to calculate an assessment score for whether the person suffers from ADHD; Interview assessment module, used to assess attention deficit or assess risk score of impulsivity / hyperactivity through interview; And, a quality of life assessment module calculates an assessment risk score based on the quality of life of the assessment subject.

6. The adult ADHD assessment system according to claim 5, characterized in that: The system also includes a feedback module for receiving the final diagnosis result and feeding back the result to the adult ADHD assessment model to adjust the parameters of the adult ADHD assessment model.

7. The adult ADHD assessment system according to claim 5, characterized in that: The method for calculating the risk score for attention deficit or impulsivity / hyperactivity includes: calculating the risk score using the Adult ADHD Diagnostic Interview 2.0 Chinese Revised Version and its standards.

8. The adult ADHD assessment system according to claim 5, characterized in that: The quality of life assessment module includes life productivity, mental health, interpersonal relationships and outlook on life assessment.

9. The adult ADHD assessment system according to claim 5, characterized in that: Each item in the quality of life assessment module is scored using a Likert scale.

Images