Early screening and intervention curative effect evaluation method for children with autism

Abstract

The invention relates to the technical field of intelligent medicine, in particular to an autism child early screening and intervention curative effect evaluation method which comprises the following steps: S1, obtaining social reaction scale data of a Chinese version; s2, acquiring and collecting data of preschool autism patients and typical development children; s3, filling missing data by adopting a mean value filling method; s4, dividing the filled data into a test set and a training set, and establishing an XGBoost model; s5, optimizing hyper-parameters in the XGBoost model by using grid search to obtain optimal model configuration; s6, Pearson correlation analysis is adopted, scale entries obtained through optimal model configuration are added to be subjected to correlation analysis with a gold standard scale capable of being used for diagnosing autism, and it is verified that the simplified scale is enough to serve as an effective tool for screening the pre-school-age autism child patients; by combining a machine learning algorithm, the screening process is optimized, and the screening efficiency and the diagnosis accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method for early screening and intervention efficacy evaluation of children with autism. Background Art

[0002] Autism Spectrum Disorder (ASD) is a pervasive neurodevelopmental disorder with the following core symptoms: ① Social interaction and communication disorders; ② Restricted interests and repetitive behaviors. In recent years, the incidence of ASD has been on the rise. According to the latest statistics, the global incidence is 1% to 2%, and there is 1 ASD patient in every 36 8-year-old children in the United States. The incidence of ASD in my country is about 1%. ASD has become the number one "killer" of mental disabilities in China.

[0003] According to surveys, about 45% of ASD patients may suffer from intellectual disability, 28-44% of ASD patients have attention deficit disorder, ≤79% of patients have abnormal motor function, and patients may also suffer from a variety of other physical, neurological and mental diseases, which imposes a huge economic and mental burden on families and society. ASD is extremely destructive to the family happiness index. Families raising ASD children face many challenges, such as: delayed diagnosis of the disease, economic and time pressures brought by long-term rehabilitation training, the impact of abnormal behavior of ASD children on the quality of family life, and various concerns about the future independent life of ASD children. Compared with normal families, in addition to coping with difficulties in the education and medical systems, the families of ASD children must also cope with behavioral and social challenges. As children enter adolescence, coupled with the lack of correct cognition and support system for adult patients in society, these pressures will gradually increase. Rehabilitation intervention for ASD is a long-term process. The lifetime economic burden of an ASD patient in the United States is 2.36 to 2.62 million US dollars, and in the United Kingdom it is 1.5 million pounds. These economic burdens include indirect and direct economic costs, from medical treatment to special education, and then to the loss of parents' labor. The average annual expenditure of ASD families in my country on education and training is 55,000 yuan, accounting for 89.9% of the family’s annual income. Early diagnosis and intervention can reduce the lifelong care cost of ASD by 2 / 3.

[0004] Currently, the average age of diagnosis for ASD is relatively old, with the median age of diagnosis being 36 to 82 months. Around 90% of children are found to be abnormal at around two years old, and 44.2% of children take more than one year from suspicion to confirmation of diagnosis. Research shows that early identification, early diagnosis, and early intervention can effectively improve the prognosis of ASD, and early treatment is beneficial to improving children's cognitive, language, and social adaptation abilities. Therefore, early identification of ASD patients is particularly important. However, the clinical symptoms and etiology of ASD are heterogeneous, and there is currently a lack of reliable standard biomarkers for early auxiliary diagnosis of ASD. The early detection and diagnosis of ASD still rely on ASD behavior assessment tools, and an effective assessment tool can play a key role in the early diagnosis of ASD. ASD usually begins before the age of 3. Children are usually diagnosed with ASD during the preschool stage when they first encounter a complex social environment, and the diagnosis becomes more stable over time. The American Academy of Pediatrics recommends conducting two ASD screenings before children reach 3 years old. The severity of social impairment at the age of 3 in children may be a predictor of the prognosis of ASD patients in adulthood.

[0005] Currently, there is still a lack of early screening tools for ASD in preschool children aged 2 to 5 in China. Considering that early identification and diagnosis of ASD are beneficial to the treatment of patients, thereby improving the social function and quality of life of ASD patients, there is an urgent need for an effective tool for screening or identifying ASD suitable for preschool children in China to facilitate the development of ASD diagnosis and intervention work in China. Therefore, the present invention provides a method for early screening and intervention efficacy evaluation of autistic children, which simplifies the traditional scale using machine learning algorithms to improve the screening efficiency and diagnostic accuracy. Summary of the Invention

[0006] To solve the above problems, the present invention provides a method for early screening and intervention efficacy evaluation of autistic children, which combines machine learning algorithms to optimize the screening process and improve the screening efficiency and diagnostic accuracy.

[0007] To achieve the above object, the technical solution of the present invention is as follows: A method for early screening and intervention efficacy evaluation of autistic children, comprising the following steps:

[0008] S1. Based on the data of the Social Responsiveness Scale in English version, after Chinese translation, the data of the Chinese version of the Social Responsiveness Scale is obtained;

[0009] S2. Obtain and collect the general information, Social Responsiveness Scale data, and gold standard scale for diagnosing autism of preschool autistic patients and typically developing children; among them, the Social Responsiveness Scale data includes 5 dimensions of social awareness, social cognition, social communication, social motivation, restricted interests, and repetitive behaviors, with a total of 65 scale items, and the evaluation scores corresponding to each scale item;

[0010] S3. Use the mean filling method to fill in the missing data of the general information and social response scale data;

[0011] S4. Divide the filled data into a test set and a training set, then use the XGBoost algorithm to model the test set to obtain the XGBoost model, and use the training set to verify the XGBoost model;

[0012] S5. Use grid search to optimize the hyperparameters in the XGBoost model, simplify the hyperparameter configuration in the XGBoost model, and obtain the best model configuration;

[0013] S6. Adopt Pearson correlation analysis, add up the scale items obtained from the best model configuration respectively to get the simplified scale; conduct correlation analysis between the simplified scale and the gold standard scale that can be used to diagnose autism.

[0014] Furthermore, in S2, the method for obtaining the evaluation score is as follows. First, add rating labels of "not in line", "somewhat in line", "often in line", and "almost always in line" based on each scale item, and then summarize based on the corresponding scores of the evaluation labels to obtain the evaluation score.

[0015] Furthermore, in S5, the hyperparameter configuration in the XGBoost model is simplified as follows: First, use feature selection to screen the importance of the scale items, and retain the scale items with non-zero importance in the XGBoost model.

[0016] Furthermore, in S5, the hyperparameters include the number of trees, learning rate, minimum weight of leaf nodes, maximum depth of trees, penalty term for splitting nodes, sampling ratio of samples when constructing each tree, sampling ratio of features when constructing each tree, control of L1 regularization term, and control of L2 regularization term.

[0017] Furthermore, in S4, before dividing the filled data into a test set and a training set, children corresponding to the behavior assessment scale will also be divided by cultural background, and the corresponding evaluation scores in different social environments, and the evaluation scores corresponding to each item will be determined based on the behavior evaluation scale in different social environments.

[0018] Furthermore, in S4, the steps for determining the evaluation scores corresponding to each scale item based on the social response scale data in different social environments are as follows: First, in different social environments, obtain the social response scale data of preschool children with autism and typically developing children, and record the social response scale data in the first social environment as the comparison table;

[0019] Next, draw a fluctuation line for the reference evaluation scores corresponding to each scale item in different social environments and the comparison evaluation scores corresponding to the comparison table. If the fluctuation lines of the comparison evaluation scores and the reference evaluation scores are stable wavy lines, output the evaluation scores of the corresponding items in the comparison table; if the fluctuation lines of the comparison evaluation scores and the reference evaluation scores are zigzag lines with large fluctuation amplitudes, obtain the subsequent social response scale data based on the chronological order of the comparison table, and output the modified evaluation scores of the corresponding scale items based on the similarity of the social response scale data.

[0020] Further, in S4, when evaluating the social response scale data in different social environments, different personnel will also be selected for separate evaluations to obtain the social response scale data in different social environments.

[0021] Further, in S1 to S6, when outputting the modified evaluation scores of the corresponding scale items based on the similarity of the social response scale data, refer to the comparison evaluation scores corresponding to the comparison table for the modified evaluation scores, mark the corresponding test set evaluation scores with the modified evaluation scores. The test set evaluation scores include the social environment corresponding to the social response scale data and the actual evaluation scores. When substituting the test set into the training model to obtain the key features, convert the actual evaluation scores into the corresponding modified evaluation scores to determine the key features.

[0022] Further, in S4, when determining the evaluation scores corresponding to each scale item based on the behavior evaluation metrics in different social environments, the evaluation scores of each scale item will also be compared for consistency in different social environments. If they are consistent, compare the similarity of the patient's answering methods under the corresponding scale item. If they are similar, remove the social response scale data; if they are not similar, perform subsequent processing on the social response scale data; if they are inconsistent, perform subsequent processing on the social response scale data.

[0023] Further, in S4, when selecting different personnel to separately evaluate the social response scale data in different social environments, different personnel will also be made to evaluate the social response scale data of the same patient in advance, compare the differences in the evaluation scores corresponding to each scale item respectively, remove the personnel corresponding to the maximum and minimum values of each scale item respectively, and evaluate the social response scale data in different social environments based on the remaining personnel.

[0024] Further, in S1 to S6, based on the best model configuration, and then based on the social response scale data under specific conditions, establish a training set and a test set under specific conditions to adjust the best model configuration, so as to establish the best model configuration under specific conditions.

[0025] The above solution has the following beneficial effects:

[0026] 1. Through data preprocessing, data partitioning, model training, feature screening, and correlation analysis, this solution effectively screens out important features in the social responsiveness scale data, optimizes the screening scale items using machine learning, simplifies the scale structure, improves the reliability and validity of screening, and translates the social responsiveness scale to make it applicable to the cultural background of Chinese children.

[0027] 2. This solution utilizes the ability of the XGBoost algorithm to handle a large number of non - linear relationships and missing values, effectively connects the data, and gradually selects key features to reduce the risks of overfitting and underfitting, and improve the prediction accuracy and reliability of the optimal model.

[0028] 3. In this solution, since the XGBoost algorithm selects key features by splitting nodes, there are a large number of associations between key features and evaluation scores in the subsequent processing. The Pearson correlation coefficient can measure the linear relationship between features and target variables, helping to identify potential key factors and providing a theoretical basis for subsequent feature selection and model interpretation.

[0029] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of an embodiment of the method for evaluating the efficacy of early screening and intervention for autistic children according to the present invention;

[0031] Figure 2 It is a schematic diagram of the data of the translated social responsiveness scale in the embodiment of the method for evaluating the efficacy of early screening and intervention for autistic children according to the present invention;

[0032] Figure 3 It is a graph of the reaction recall rate and precision data in the embodiment of the method for evaluating the efficacy of early screening and intervention for autistic children according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The technical solutions of the present invention will be described clearly and completely below with reference to the accompanying drawings. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] The following is a further detailed description through specific embodiments:

[0035] Example 1:

[0036] As shown in the atta Figures 1 to 2Shown: A method for early screening and intervention efficacy evaluation of autistic children, comprising the following steps:

[0037] S1. Based on the data of the Social Responsiveness Scale in English version, after Chinese translation, the data of the Chinese version of the Social Responsiveness Scale is obtained. The specific steps are as Figure 2 shown.

[0038] S2. Obtain and collect the general information, Social Responsiveness Scale data of preschool autistic patients and typically developing children, and the gold standard scale for diagnosing autism; among them, the Social Responsiveness Scale data includes 5 dimensions of social awareness, social cognition, social communication, social motivation, restricted interests and repetitive behaviors, with a total of 65 scale items, and the evaluation scores corresponding to each scale item; among them, the method for obtaining the evaluation scores is as follows: first, rating labels of "not in line", "somewhat in line", "often in line" and "almost always in line" are added based on each scale item, and then the evaluation scores are obtained by summarizing the corresponding scores of the evaluation labels.

[0039] S3. Use the mean filling method to fill in the missing data of the general information and Social Responsiveness Scale data; since mean filling is a commonly used and effective method for handling missing values, it can reduce the deviation of data and avoid artificially introducing extreme values. For data, missing values may be caused by incomplete collection or other uncontrollable factors. Mean filling can effectively maintain the stable distribution of data, reduce unnecessary fluctuations, and help improve the robustness of the model.

[0040] S4. Divide the filled data into a test set and a training set, then use the XGBoost algorithm to build a model for the test set to obtain an XGBoost model, and use the training set to verify the XGBoost model; among them, the feature importance provided by XGBoost is measured by the split nodes in the tree structure, and the features with higher importance contribute more to the prediction ability of the model. Through this screening process, we effectively reduce redundant features and improve the interpretability and computational efficiency of the model.

[0041] Meanwhile, since the XGBoost algorithm is an efficient decision tree algorithm based on gradient boosting, during the processing, a training model is established by optimizing hyperparameters. Among them, the number of trees may better fit the data by building more trees, but it also increases the risk of overfitting; the learning rate can control the contribution of each tree to the final prediction and prevent the model from being too complex; the minimum weight of the leaf nodes can avoid overfitting; the maximum depth of the tree can limit the depth to prevent overfitting; the higher the penalty term for splitting nodes is set, the stricter it is, thus controlling the complexity of the tree; controlling the sampling ratio of samples when building each tree helps reduce variance and improve generalization performance; controlling the sampling ratio of features when building each tree reduces the model complexity; by controlling the L1 and L2 regularization terms respectively, irrelevant features are suppressed and the model stability is improved.

[0042] S5. Use grid search to optimize the hyperparameters in the XGBoost model, and then adopt feature selection to screen the importance of scale items, and retain the scale items with non-zero importance in the XGBoost model to simplify the hyperparameter configuration in the XGBoost model and obtain the best model configuration; among them, the hyperparameters include: n_estimators (the number of trees, more trees may better fit the data, but it also increases the risk of overfitting), learning_rate (the learning rate, controlling the contribution of each tree to the final prediction and preventing the model from being too complex), min_child_weight (the minimum weight of the leaf nodes, avoiding overfitting), max_depth (the maximum depth of the tree, limiting the depth can prevent overfitting), gamma (the penalty term for splitting nodes, the higher the value, the stricter it is, thus controlling the complexity of the tree), subsample (controlling the sampling ratio of samples when building each tree, helping to reduce variance and improve generalization performance), colsample_bytree (controlling the sampling ratio of features when building each tree, reducing the model complexity), reg_alpha and reg_lambda (controlling the L1 and L2 regularization terms respectively, suppressing irrelevant features and improving the model stability).

[0043] S6. Adopt Pearson correlation analysis, add up the scale items obtained from the best model configuration respectively to get the simplified scale; conduct correlation analysis between the simplified scale and the gold standard scale that can be used to diagnose autism, so as to verify that the simplified scale is sufficient to be used as an effective tool for screening preschool children with autism.

[0044] Among them, the Pearson correlation coefficient can measure the linear relationship between features and the target variable, help identify potential key factors, provide a theoretical basis for subsequent feature selection and model interpretation, so as to use the best model to screen and evaluate the intervention efficacy of subsequent children.

[0045] In this embodiment, 648 ASD children aged 2 - 5 years from 3 Class III Grade A hospitals and 3 special education schools in Hunan, Guangdong, Shandong and other places, and 138 typically developing children who attended ordinary schools or had health check-ups at the health check-up center of a certain hospital during the same period were studied. General information of the research subjects, data of the Social Responsiveness Scale, and behavioral data reflecting the gold standard for ASD diagnosis were collected as basic data for verification.

[0046] (1) Use the mean filling method to fill in the missing data;

[0047] First, check each variable (such as each item in the SRS scale) to find all missing values. Then, for each feature with missing values, calculate the mean of the non-missing data of this feature. Finally, fill in the missing values with the mean of this feature to ensure that the data after filling can maintain the mean of this variable unchanged, thereby reducing the bias caused by missing data.

[0048] (2) Randomly divide the data into a test set and a training set. Specifically: the sample size of ASD children in the test set is 194, and the sample size of typically developing children is 40; the sample size of ASD children in the training set is 454, and the sample size of typically developing children is 98.

[0049] (3) Use the XGBoost algorithm for modeling;

[0050] First, use the XGBoost algorithm for modeling. The specific process includes: 1. Data preparation: Format the features (including gender, age, SRS scale items, etc.) and target variables (ASD or typical development) in the training set to ensure that the data structure is suitable for the XGBoost model. 2. Model initialization: Import the XGBoost library and create a basic XGBoost classification model. 3. Model training: Use the training set to train the model. The code is as follows:

[0051]

[0052]

[0053]

[0054]

[0055] (4) Use grid search to optimize the hyperparameters after modeling to obtain the best model configuration;

[0056] By using the grid search algorithm and 5-fold cross-validation, the final parameters of the XGBoost model were determined. The optimal parameters obtained after training are 'learning_rate': 0.1, 'n_estimators': 50,'subsample': 0.2, 'colsample_bytree': 0.2,'max_depth': 2,'min_child_weight': 2,'reg_alpha': 0.15,'reg_lambda': 0.15, 'gamma': 0.09.

[0057] (5) Feature selection was used to screen the importance of the scale items, and the features (i.e., scale items) with non-zero importance in the XGBoost model were retained;

[0058] Among them, the input features F = [gender, age, SRS1, SRS2, SRS3, SRS4,..., SRS65],

[0059] Features with non-zero importance =

[0060] [gender, SRS3, SRS5, SRS6, SRS8, SRS10, SRS11, SRS12, SRS13, SRS14,

[0061] SRS15, SRS16, SRS18, SRS22, SRS23, SRS24, SRS25, SRS26, SRS29, SRS30,

[0062] SRS31, SRS32, SRS33, SRS35, SRS37, SRS38, SRS43, SRS44, SRS47, SRS48,

[0063] SRS52, SRS56, SRS63, SRS65]

[0064] (6) A simplified version of the Social Responsiveness Scale assessment tool was formed based on step 5;

[0065] (7) Pearson correlation analysis was used to show that the simplified Social Responsiveness Scale assessment tool is correlated with the behavioral data reflecting the gold standard for ASD diagnosis (as shown in Figure 3 ).

[0066] Example 2:

[0067] The difference from Example 1 is that in S4, before dividing the filled data into a test set and a training set, children corresponding to the behavioral assessment scale will also be divided according to their cultural backgrounds, as well as the corresponding assessment scores in different social environments. Based on the behavioral evaluation scale in different social environments, the assessment scores corresponding to each item are determined. In different social environments, the assessment scores of each scale item will also be compared for consistency. If they are consistent, the answering methods of patients under the corresponding scale items will be compared for similarity. If they are similar, the data of the social response scale will be removed; if they are not similar, the data of the social response scale will be processed subsequently; if they are inconsistent, the data of the social response scale will be processed subsequently.

[0068] For example, due to a large number of cognitive differences in different cultural backgrounds, adaptive adjustments are made based on the cultural backgrounds of children to ensure the accuracy and effectiveness of the evaluation results. At the same time, the poor performance of individual patients in certain social environments may be misdiagnosed as social disorders, while in fact it is due to the temporary social difficulties of the patients. Through behavioral evaluations in different social environments, more social support can be provided to patients to ensure the accuracy and effectiveness of the evaluation results.

[0069] By excluding the interference with similarities in stereotyped behaviors and social communication, since the patient is a child, during the measurement of the social response scale data, due to the development of the child's own intelligence, there may be a large number of stereotyped behaviors, which interfere with the subsequent assessment of social disorders. In this way, it is possible to distinguish the difficulty in identifying intellectual disability and poor social ability in the social response scale data.

[0070] The steps to determine the assessment scores corresponding to each scale item based on the behavioral evaluation scale in different social environments are as follows: First, in different social environments, obtain the social response scale data of children with autism and typically developing children. Denote the social response scale data in the first social environment as the comparison table;

[0071] Then, draw a fluctuation line for the reference assessment scores corresponding to each scale item in different social environments and the comparison assessment scores corresponding to the comparison table. If the fluctuation lines of the comparison assessment scores and the reference assessment scores are smooth wavy lines, the assessment scores of the corresponding scale items in the comparison table are output; if the fluctuation lines of the comparison assessment scores and the reference assessment scores are zigzag lines with large fluctuation amplitudes, obtain the subsequent social response scale data based on the time sequence of the comparison table, and output the modified assessment scores of the corresponding scale items based on the similarity of the social response scale data.

[0072] By comparing the possible differences in the evaluation scores in different social environments, the evaluation scores in different social environments are obtained. Through comparison, the evaluation scores corresponding to each scale item are corrected to make the data in the model establishment process more accurate, so as to ensure the prediction accuracy of the model establishment.

[0073] Embodiment 3:

[0074] The difference from Embodiment 2 is that in S4, when evaluating the social response scale data in different social environments, different personnel will also be selected for separate evaluation to obtain the social response scale data in different social environments. At the same time, different personnel will be made to evaluate the social response scale data of the same patient in advance, and the differences in the evaluation scores corresponding to each scale item will be compared separately. The personnel corresponding to the maximum and minimum values of each scale item will be removed respectively, and the social response scale data will be evaluated in different social environments based on the remaining personnel.

[0075] For example, by using different personnel to reduce the subjective judgment of the evaluator, so as to ensure the objectivity of the evaluation scores of the social response scale data in different social environments, which is convenient for subsequent correction of the evaluation scores corresponding to each scale item. By selecting personnel, while reducing the subjectivity in the evaluation process, the consistency of the measurement results is ensured, and thus the accuracy and effectiveness of the evaluation results are improved.

[0076] In S1 to S6, when the modified evaluation scores corresponding to the scale items are output based on the similarity of the social response scale data, the modified evaluation scores are referenced to the comparison evaluation scores in the comparison table, and the corresponding test set evaluation scores are marked with the modified evaluation scores. The test set evaluation scores include the social environment corresponding to the social response scale data and the actual evaluation scores. When the test set is substituted into the training model to obtain the key features, the actual evaluation scores are converted into the corresponding modified evaluation scores to determine the key features.

[0077] At the same time, a training model is established based on the modified evaluation scores in the training process. The data used is somewhat different from the actual data. In the subsequent verification process through the test set, the associated features are obtained based on the marking to reduce the data gap, and thus the accuracy of screening the key features is improved.

[0078] Embodiment 4:

[0079] The difference from Embodiment 3 is that in S1 to S6, based on the best model configuration, and then based on the social response scale data under specific conditions, a training set and a test set under specific conditions are established to adjust the best model configuration, so as to establish the best model configuration under specific conditions.

[0080] For example, social assessment scales in different individuals or social environments often have specific individuals. The optimal model is adjusted through experimental data under different conditions so that the optimal model under specific conditions can perform atopic recognition on some patients, improve the recognition accuracy for different patients, reduce the dependence on the standardized optimal model established under standardized scenarios, and reduce the limitations of the standardized optimal model in recognition.

[0081] Obviously, the above-mentioned embodiments are merely examples given for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or alterations can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. And the obvious changes or alterations derived therefrom still fall within the protection scope of the present invention.

Claims

1. A method for early screening and intervention efficacy evaluation of children with autism, characterized in that: The following steps are involved: S1. Based on the English version of the social response scale data, after Chinese translation, the Chinese version of the social response scale data was obtained; S2. Obtain and collect general information of preschool autistic patients and typically developing children, social responsiveness scale data, and gold standard scales that can be used to diagnose autism; the social responsiveness scale data includes five dimensions of social awareness, social cognition, social communication, social motivation, restricted interests, and repetitive behaviors, a total of 65 scale items, and the corresponding assessment scores of each scale item; S3, use mean filling method to fill in the missing data of general information and social response scale data; S4, divide the filled data into a test set and a training set, then use the XGBoost algorithm to model the test set to obtain the XGBoost model, and use the training set to verify the XGBoost model; S5. Use grid search to optimize the hyperparameters in the XGBoost model, simplify the hyperparameter configuration in the XGBoost model, and obtain the best model configuration; S6. Using Pearson correlation analysis, the scale items obtained from the best model configuration were added together to obtain a simplified scale; the simplified scale was correlated with the gold standard scale that can be used to diagnose autism.

2. The method for early screening and intervention efficacy evaluation of children with autism according to claim 1, characterized in that: In S2, the method for obtaining the evaluation score is as follows: first, based on each scale item, a rating label of not meeting, somewhat meeting, often meeting, and almost always meeting is added, and then based on the corresponding scores of the evaluation labels, the evaluation score is obtained by summarizing them.

3. The method for early screening and intervention efficacy evaluation of children with autism according to claim 2, characterized in that: In S5, the hyperparameter configuration in the simplified XGBoost model is as follows: First, feature selection is used to screen the importance of scale items, and the scale items whose importance in the XGBoost model is not 0 are retained.

4. The method for early screening and intervention efficacy evaluation of autistic children according to claim 3, characterized in that: In S5, the hyperparameters include the number of trees, the learning rate, the minimum weight of the leaf nodes, the maximum depth of the tree, the penalty term for splitting nodes, the sampling ratio of samples when each tree is built, the sampling ratio of features when each tree is built, the control L1 regularization term, and the control L2 regularization term.

5. The method for early screening and intervention efficacy evaluation of autistic children according to claim 4, characterized in that: In S4, before dividing the filled data into test and training sets, the children corresponding to the behavior assessment scale are divided into cultural backgrounds, and the corresponding assessment scores in different social environments are determined based on the behavior assessment scale in different social environments to determine the assessment scores corresponding to each item.

6. The method for early screening and intervention efficacy evaluation of autistic children according to claim 5, characterized in that: In S4, the steps for determining the evaluation scores corresponding to each scale item based on the social response scale data in different social environments are as follows: first, the social response scale data of preschool autistic patients and typically developing children are obtained in different social environments, and the social response scale data in the first social environment are recorded as the comparison table; Then, a fluctuation line is drawn between the reference evaluation score corresponding to each scale item in different social environments and the comparison evaluation score corresponding to the comparison table. If the fluctuation line of the comparison evaluation score and the reference evaluation score is a smooth wavy line, the evaluation score of the corresponding item in the comparison table is output; if the fluctuation line of the comparison evaluation score and the reference evaluation score is a zigzag line with a large fluctuation amplitude, the subsequent social reaction scale data is obtained based on the time sequence of the comparison table, and the modified evaluation score of the corresponding scale item is output based on the similarity of the social reaction scale data.

7. The method for early screening and intervention efficacy evaluation of autistic children according to claim 6, characterized in that: In S4, the social response scale data is evaluated in different social environments, and different personnel are selected for separate evaluation to obtain the social response scale data in different social environments.

8. The method for early screening and intervention efficacy evaluation of autistic children according to claim 7, characterized in that: In S1 to S6, when the modified evaluation score of the corresponding scale item is output based on the similarity output of the social reaction scale data, the modified evaluation score is referenced with the comparison evaluation score corresponding to the comparison table, and the corresponding test set evaluation score is marked with the modified evaluation score. The test set evaluation score includes the social environment and actual evaluation score corresponding to the social reaction scale data. When the test set is substituted into the training model to obtain the key features, the actual evaluation score is converted into the corresponding modified evaluation score to determine the key features.

9. The method for early screening and intervention efficacy evaluation of children with autism according to claim 8, characterized in that: In S4, when determining the evaluation scores corresponding to each scale item based on the behavioral evaluation indicators in different social environments, the evaluation scores of each scale item will be compared for consistency in different social environments. If they are consistent, the patient's answer methods under the corresponding scale items will be compared for similarity. If they are similar, the social reaction scale data will be removed; if they are not similar, the social reaction scale data will be processed subsequently; if they are inconsistent, the social reaction scale data will be processed subsequently.

10. The method for early screening and intervention efficacy evaluation of autistic children according to claim 9, characterized in that: In S4, when different personnel are selected to evaluate the social response scale data in different social environments, different personnel will also evaluate the social response scale data of the same patient in advance, compare the differences in the evaluation scores corresponding to each scale item, remove the personnel corresponding to the maximum and minimum values ​​of each scale item, and evaluate the social response scale data in different social environments based on the retained personnel.

11. The method for early screening and intervention efficacy evaluation of autistic children according to claim 10, characterized in that: In S1 to S6, based on the best model configuration, the best model configuration is adjusted based on the social response scale data under specific conditions to establish a training set and a test set under specific conditions, so as to establish the best model configuration under specific conditions.

Images