System for aiding in assessing autism spectrum disorder

Abstract

A system for aiding in assessing autism spectrum is adaptive for solving the problem of lacking scientific quantitative indicators for assessing autism tendencies in conventional ways. The system includes an image displaying device, an EEG measurement device, a processing module and a trained assessment model. The image displaying device is configured to display a scenario-based video with a question. The EEG measurement device is configured to acquire brainwave signals of an assessed subject after watching the scenario-based video and the question. The processing module controls the image displaying device to display the scenario-based video with the question and records the EEG signals obtained by the EEG measurement device. The trained assessment model has a correlating relationship between an input data and an autism tendency of the observed subject. Based on the correlating relationship, the assessment model generates an evaluating index to assist in assessing the autism tendency of the subject when receiving the input data. This invention achieves the effect of generating auxiliary assessments through scientific quantitative indicators.

Description

Technical Field

[0001] The present invention relates to symptom assessment technology, in particular to an auxiliary assessment system for autism. Prior Art

[0002] Conventional assessment methods for autism / Autism Spectrum Disorder (ASD) rely on subjective evaluations of the individual's performance or status across a variety of specific indicators by primary caregivers (e.g., parents or teachers), and on physicians' personal experience to conduct assessments and diagnoses. However, these assessment methods involve subjective judgments by caregivers and physicians, and different caregivers and physicians often make different assessments. This results in a lack of standardized, quantifiable basis for autism assessments, making it difficult to effectively determine whether a child possesses autistic traits.

[0003] In view of this, conventional autism assessment methods still need to be improved. Summary of the Invention

[0004] To solve the above problems, the purpose of the present invention is to provide an auxiliary assessment system for autism, which can perform auxiliary assessment based on scientific quantitative indicators.

[0005] The use of the quantifiers "a" or "an" in the elements and components described throughout the present invention is merely for convenience and to provide a general meaning of the scope of the present invention; in the present invention, they should be interpreted as including one or at least one, and the single concept also includes the plural case, unless it is obvious that it means otherwise.

[0006] The term "coupling" as used throughout the present invention includes direct or indirect electrical and / or signal connections, which can be selected by those skilled in the art based on usage requirements.

[0007] The technical features of the "processing module" described throughout the present invention include at least one processor and corresponding non-volatile memory to implement corresponding data processing, calculation and storage.

[0008] The auxiliary assessment system for autism of the present invention comprises an image playback device for playing a situational image with a question, so that the subject being evaluated can think about the question after watching the situational image; an electroencephalogram (EEG) measuring device for obtaining the electroencephalogram (EEG) signal of the subject being evaluated after watching the situational image and the question; and a processing module coupled to the image playback device and the electroencephalogram (EEG) measuring device respectively to control the image playback device to play the situational image with the question, and to record the electroencephalogram (EEG) signal obtained by the electroencephalogram (EEG) measuring device in a preliminary recording time interval starting from a questioning time point no later than the question being presented to the image playback device; the processing module comprises: a processing and storage unit for storing the electroencephalogram (EEG) signal in the preliminary recording time interval, and defining the electroencephalogram (EEG) signal as an event. The invention relates to an event-related potential information; and a trained evaluation model having a correlation between input data and a subject's autism tendency, wherein the evaluation model generates an evaluation index based on the correlation when receiving the input data to assist in evaluating the subject's autism tendency; the input data is one of a first set of data, a second set of data, and a third set of data; the first set of data has the event-related potential information and event-related spectrum disturbance information; the second set of data has functional connection information; the third set of data has the event-related potential information, the event-related spectrum disturbance information, and the functional connection information; the processing and storage unit converts the event-related potential information into the event-related spectrum disturbance information through Morlet wavelet, and converts the event-related spectrum disturbance information into the functional connection information through phase delay index.

[0009] Accordingly, the auxiliary assessment system for autism of the present invention can achieve the effect of generating auxiliary assessments using scientific quantitative indicators through the assessment model having the correlation between the input data and the autism tendency of the observed subject, and through the correlation between the input data and the brain wave signal.

[0010] During the training process, the evaluation model inputs multiple training data into multiple learning models. From the output results of each learning model, the training data and the learning model corresponding to the one with the highest accuracy are selected as the input data and the evaluation model, respectively. The training data includes multiple typical cases and multiple atypical cases, and the multiple typical case data and multiple atypical case data are generated after the question is asked in the situational image. The types of training data corresponding to the multiple typical case data and the multiple atypical case data include the first set of data, the second set of data, and the third set of data. Thus, by selecting the training data and the learning model with the highest accuracy during the training process as the input data and the evaluation model, respectively, the accuracy of the evaluation model is improved.

[0011] The classifiers of the learning model include at least two of a KNN classifier, a BDT classifier, a GAB classifier, an SVM classifier using a sigmoid kernel function, an SVM classifier using an RBF kernel function, and an SVM classifier using a polynomial kernel function. Thus, by selecting the learning model with the highest accuracy as the evaluation model during training, the accuracy of the evaluation model is improved.

[0012] The training data comprises a plurality of sensors arranged at different locations according to the international 10-20 system of electroencephalography, wherein the locations include at least P3, P4, P7, and Pz, and each sensor corresponding to each location is defined as sensor P3, sensor P4, sensor P7, and sensor Pz; the event-related potential information in the first set of data in the training data includes information about sensor P7 within 160 to 220 ms after the event trigger point, and includes information about sensor Pz within 400 to 500 ms after the event trigger point; the event-related spectral disturbance information in the first set of data in the training data includes information about sensor P3 within 215 to 280 ms after the event trigger point and within a frequency range of 3 to 5 Hz, and includes information about sensors P4 and Pz within 245 to 320 ms after the event trigger point and within a frequency range of 4 to 6 Hz. In this way, through specific training data, especially training data that is significantly different from typical case data, it is possible to improve the accuracy of the training model and the evaluation model.

[0013] The functional connection information in the second set of data used as training data includes information between 800 and 900 ms at a frequency of 4 to 7 Hz, and information between 300 to 400 ms, 400 to 500 ms, and 800 to 900 ms at a frequency of 8 to 12 Hz. This allows for improved accuracy of both the training and evaluation models through the use of specialized training data, particularly data that differs significantly from typical case data.

[0014] The training data is a plurality of sensors arranged at different positions according to the international 10-20 system of electroencephalography, wherein the positions include P3, P4, P7, and Pz, and each sensor corresponding to each position is defined as a sensor P3, a sensor P4, a sensor P7, and a sensor Pz respectively; the event-related potential information in the third set of data in the training data includes information of the sensor P7 from 160 to 220 ms after the event trigger point, and includes information of the sensor Pz from 400 to 500 ms after the event trigger point; the event-related spectral disturbance information in the third set of data in the training data includes information of the sensor P3 from 215 to 280 ms after the event trigger point and between frequencies of 3 and 5 Hz, and includes information of the sensors P4 and Pz from 245 to 320 ms after the event trigger point and between frequencies of 4 and 6 Hz; the functional connection information in the third set of data in the training data includes information between 800-900 ms and frequencies between 4-7 Hz, and information between 300-400 ms, 400-500 ms, and 800-900 ms and frequencies between 8-12 Hz. In this way, using specific training data, especially training data that differs significantly from typical case data, can improve the accuracy of training and evaluation models.

[0015] The evaluation model's classifier uses an SVM classifier with a sigmoid kernel function, and the input data is the first set of data. This improves the accuracy of the evaluation model by selecting the training data and learning model with the highest accuracy during training as the input data and evaluation model, respectively. Simple diagram description

[0016] [Figure 1] A block diagram of the system for assisting autism assessment according to the present invention. [Figure 2] Schematic diagram of a situational image with questions according to the progression of time on the timeline. Implementation Method

[0017] In order to make the above and other objects, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments of the present invention are given below and described in detail with reference to the accompanying drawings.

[0018] Referring to FIG. 1 , a block diagram of the autism assessment system according to the present invention is shown. The autism assessment system includes an image playback device 1, an EEG measurement device 2, and a processing module 3. The image playback device 1 and the EEG measurement device 2 are each coupled to the processing module 3.

[0019] The image playback device 1 is used to play a situational video with a question, so that the evaluated subject can view the situational video and reflect on the question. The situational video specifically includes content related to interpersonal interaction, such as simply presenting a person's expression or depicting the process of a specific event. The question can ask a question corresponding to the situational video, such as inquiring about the inner feelings reflected by the person's expression in the video, or inquiring about the person's mood, thoughts, or appropriate response after experiencing a specific event. In particular, the content presented to the observed subject by the image playback device can include content from a social cognitive interaction experimental paradigm.

[0020] The EEG measurement device 2 is used to acquire / detect changes in the subject's brainwaves, particularly the subject's brainwave signals after viewing the situational image and receiving the question. The brainwave signals may be EEG (electroencephalography) signals. In one example, the EEG measurement device 2 may include multiple sensors / measuring electrodes to acquire corresponding brain signals from multiple locations on the subject's brain. Alternatively, the EEG measurement device 2 may be a universal EEG device with sensors configured according to the International 10-20 EEG system (F3, F4, F7, F8, Fz, C3, C4, P3, P4, P7, P8, and Pz) corresponding to different scalp locations. Preferably, the reference electrode A1 is located on the subject's left earlobe, and the corresponding EEG signals are amplified and recorded at a sampling rate of 500 Hz. In particular, each sensor's configuration position has a corresponding position brainwave signal, and the brainwave signal has multiple position brainwave signals.

[0021] It should be noted that the EEG measurement device 2 is understandable to those with ordinary knowledge in the field of the present invention, and the location and number of the specific devices and sensors described above are not intended to limit the present invention. In particular, in the future, mechanisms or devices that can measure EEG signals through non-contact measurement are also included in the scope defined by the purpose of obtaining EEG signals in the present invention.

[0022] The processing module 3 is coupled to the image playback device 1 and the brainwave measurement device 2, respectively, to control the image playback device 1 to play the situational image containing the question and to record the brainwave signals acquired by the brainwave measurement device 2 during a preliminary recording time interval, starting no later than the questioning time point at which the question is presented, with the preliminary recording time interval ending after the questioning time point. Thus, by recording the corresponding brainwave signals during the specific time interval, the processing module 3 can record changes in the subject's brainwave signals after the question is asked.

[0023] To facilitate understanding of the aforementioned concepts regarding "situational image," "question," "question timing," and "initial recording time interval," Figure 2 is used as an example. The nouns representing each characteristic, content, or action are directly labeled in Figure 2. Specifically, the situational image shown in Figure 2 changes between three scene frames and contains two questions. In the first scene frame, the situation is explained; at this time, brainwave signals may or may not be recorded during this time interval. In the second scene frame, the first question is asked at the time of the first question. At this time, the corresponding brainwave signals should be recorded for a first specific time interval, starting no later than the first question timing. This first specific time interval should also allow the subject to think and react. In the third scene frame, the second question is asked at the time of the second question. At this time, the corresponding brainwave signals should be recorded for a second specific time interval, starting no later than the second question timing. This second specific time interval should also allow the subject to think and react.

[0024] Specifically, the processing module 3 includes a processing and storage unit 31 and a trained evaluation model 32. The processing and storage unit 31 is used to store the aforementioned EEG information that has a specific temporal correlation with the scene image. In particular, since the EEG information is acquired after the corresponding question is asked, and the corresponding question can be considered an event trigger, the EEG information can be correspondingly defined as a set of event-related potential (ERP) information. Optionally, a noise removal step can be performed to remove noise from the EEG information / ERP. In one example, brain signals and noise can be distinguished based on artifact subspace reconstruction (ASR) using independent component analysis (ICA).

[0025] The processing and storage unit 31 can convert the time-domain ERP information, preferably the ERP information obtained through the noise removal step, into a set of frequency-domain ERP information and / or functional connectivity information. Specifically, the ERP information is calculated using the Morlet wavelet transform method, and the functional connectivity information is obtained using the Phase Lag Index (PLI).

[0026] The trained assessment model 32 has a correlation between input data and the subject's autism propensity. Based on this correlation, upon receiving the input data, the assessment model 32 generates an assessment indicator to assist in assessing the subject's autism propensity. Specifically, the assessment model 32 is a model established through a training process using machine learning techniques and is coupled to the processing and storage unit 31 to receive at least one of the event-related potential information, the event-related spectral perturbation information, and the functional connectivity information as input data. For example, the assessment indicator can be a probability; a high probability indicates a higher autism propensity, while a low probability indicates a lower autism propensity.

[0027] Specifically, during the training process of the evaluation model 32, multiple training data are input into various learning models M, and the learning model M with the highest accuracy is selected as the trained evaluation model 32. The multiple training data include multiple typical individuals and multiple atypical individuals. The typical and atypical individuals are generated after the contextual image is asked questions. Each of the typical and atypical individuals includes corresponding event-related potential information, event-related spectral perturbation information, and functional connectivity information. In particular, the atypical individuals are individuals diagnosed with autism by a physician. Thus, by inputting the typical and atypical individual data into the learning model M corresponding to the evaluation model 32 to be trained, and after learning and verification, the evaluation performance of various learning models M can be obtained. The evaluation performance includes sensitivity, specificity, and accuracy, with accuracy preferably being used to determine the appropriate learning model M for the evaluation model 32.

[0028] In particular, by adjusting the data combination (type) in the training data, it is also possible to obtain an evaluation of the performance of the association between various input data and various learning models M. In one example, the data combination includes a first data combination, a second data combination, and a third data combination. The first data combination is composed of event-related potential information and event-related spectral perturbation information, representing information composed of local brain signals. The second data combination is composed of functional connectivity information, representing information composed of global brain signals. The third data combination is composed of event-related potential information, event-related spectral perturbation information, and functional connectivity information, representing integrated brain signal information. Thus, in a preferred example, a training data / data combination and a learning model M with the highest accuracy output are selected as the input data and selected model for the evaluation model 32.

[0029] Specifically, each learning model M has a different learning network architecture or a different classifier. When the learning network architecture is the same, different hyperparameter configurations or different loss functions can be considered different learning models M. When the classifier is the same, different computational methods can be considered different learning models M.

[0030] In one example, the multiple learning models M include a KNN (K Nearest Neighbor) classifier, a BDT (Binary Decision Tree) classifier, a GAB (Gentle Adaptive Boosting) classifier, an SVM (Support Vector Machine) classifier using a sigmoid kernel function, an SVM classifier using an RBF (Radial Based Function) kernel function, and an SVM classifier using a polynomial kernel function.

[0031] Based on the aforementioned multiple learning model M and data combinations, Table 1 below presents the training results of different learning model M and different data combinations.

[0032] Table 1: Training results of learning model M with different data combinations. [Classifier] [Input information] [Sensitivity] [Specificity] [Accuracy] 1 KNN First data set 100.0% 66.7% 83.3% Second data set 75.0% 50.0% 62.5% The third data combination 91.7% 66.7% 79.2% 2 BDT First data set 75.0% 66.7% 70.8% Second data set 66.7% 41.7% 54.2% The third data combination 58.3% 66.7% 62.5% 3 GAB First data set 83.3% 91.7% 87.5% Second data set 41.7% 66.7% 54.2% The third data combination 75.0% 58.3% 66.7% 4 Support Vector Machine Using kernel functions Sigmoid First data set [100.0 %] [91.7%] [95.8%] Second data set 83.3% 75.0% 79.2% The third data combination 83.3% 100.0% 91.7% 5 Support Vector Machine Using kernel functions RBF First data set 91.7% 91.7% 91.7% Second data set 58.3% 83.3% 70.8% The third data combination 83.3% 83.3% 83.3% 6 Support Vector Machine Using kernel functions polynomial First data set 91.7% 83.3% 87.5% Second data set 75.0% 58.3% 66.7% The third data combination 91.7% 83.3% 87.5%

[0033] As shown in Table 1, the optimal learning model M for assisting autism assessment in specific contextual images is the SVM classifier with a sigmoid kernel function. The optimal input data is the first data combination (consisting of event-related potential information and event-related spectral perturbation information), achieving the highest accuracy of 95.8%, a sensitivity of 100.0%, and a specificity of 91.7%.

[0034] Preferably, after obtaining the event-related potential information, event-related spectral perturbation information, and functional connectivity information, a difference and significance analysis is performed between typical case data and atypical case data to screen out highly relevant information that can be used to assist in the assessment of autism. This information is then used as training data for the learning model M or as input data for the assessment model 32. The difference and significance analysis is particularly performed using analysis of variance (ANOVA), but the present invention is not limited to this method.

[0035] The highly correlated information in the event-related potential information includes at least one of the following information: (1) Sensor P3 information from 400 to 500 ms after the event trigger point. (2) Sensor P4 information from 400 to 500 ms after the event trigger point. (3) Sensor P7 information from 160 to 220 ms and 400 to 500 ms after the event trigger point. In particular, the p-value for the information from 160 to 220 ms is 0.030. (4) Sensor P8 information from 400 to 500 ms after the event trigger point. (5) Sensor Pz information between 400-500 ms and 550-750 ms after the event trigger point. In particular, the p-value for the information between 400-500 ms is 0.035.

[0036] The highly correlated information in the event-related spectral disturbance information includes at least one of the following information: (1) Sensor P3 collects information from 215 to 280 ms after the event trigger, preferably in the frequency range of 3 to 5 Hz. Specifically, when the frequency is in the 3 to 5 Hz range and the time is 232 to 274 ms, the corresponding p-value is 0.046. (2) Sensor P4 data from 245 to 320 ms after the event trigger point, preferably in the frequency range of 4 to 6 Hz. Specifically, when the frequency is in the 4 to 6 Hz range and the time is 260 to 304 ms, the corresponding p-value is 0.023. (3) Sensor Pz information from 245 to 320 ms after the event trigger point, preferably in the frequency range of 4 to 6 Hz. In particular, when the frequency is in the 4 to 6 Hz range and the time is 260 to 304 ms, the corresponding p-value is 0.023. (4) Sensor Pz information from 235 to 290 ms after the event trigger point, preferably in the 1 to 2 Hz frequency range. Preferably, the time is 250 to 274 ms3.

[0037] The highly relevant information in the functional link information includes at least one of the following information: (1) 800-900 ms information at 4-7 Hz. (2) Information of 300-400 ms, 400-500 ms, and 800-900 ms at 8-12 Hz.

[0038] It should be noted that while the above content utilizes event-related potential information and / or event-related spectral perturbation information as input, training, and analysis information, information derived from event-related potential information and / or event-related spectral perturbation information through any other data processing method is also within the scope of the present invention. In other words, when referring to event-related potential information, it includes information derived from event-related potential information; when referring to event-related spectral perturbation information, it includes information derived from event-related spectral perturbation information.

[0039] In summary, the autism assessment system of the present invention utilizes an assessment model that correlates input data with a subject's autism propensity. Upon receiving the input data, it generates corresponding assessment indicators, thereby assisting in assessing the subject's autism propensity. Furthermore, during the training process, the accuracy of the assessment model can be improved by selecting the training data and learning model with the highest accuracy as input data and the assessment model, respectively. Furthermore, the accuracy of both the training model and the assessment model can be enhanced by using specific training data and input data.

[0040] While the present invention has been disclosed using the preferred embodiments described above, they are not intended to limit the present invention. Any modifications and variations made by persons skilled in the art without departing from the spirit and scope of the present invention are still within the technical scope protected by the present invention. Therefore, the scope of protection of the present invention encompasses all variations within the meaning and scope of equivalents set forth in the appended claims. Furthermore, if the aforementioned optional embodiments can be combined, the present invention encompasses any combination of such variations.

[0041] 1: Video playback device 2: Brainwave measurement device 3: Processing module 31: Processing and storage unit 32: Evaluate the model

Claims

1. An auxiliary assessment system for autism, comprising: an image playback device for playing a situational image containing a question, for a subject to be evaluated to think about the question after viewing the situational image; an electroencephalogram (EEG) measurement device for acquiring an electroencephalogram (EEG) signal from the subject after viewing the situational image and the question; and a processing module, coupled to the image playback device and the electroencephalogram (EEG) measurement device, for controlling the image playback device to play the situational image containing the question and recording the electroencephalogram (EEG) signal acquired by the electroencephalogram (EEG) measurement device within a preliminary recording time interval, starting at a questioning time point no later than when the question is presented to the image playback device. The processing module comprises: A processing and storage unit for storing the EEG signal in the preliminary recording time interval, wherein the EEG signal is defined as event-related potential information; and a trained evaluation model having a correlation between input data and the autism tendency of the observed subject, wherein the evaluation model generates an evaluation index to assist in evaluating the autism tendency of the observed subject when receiving the input data based on the correlation; the input data is one of a first set of data, a second set of data, and a third set of data; the first set of data has the event-related potential information and event-related spectral disturbance information; the second set of data has functional connection information; the third set of data has the event-related potential information, the event-related spectral disturbance information, and the functional connection information; the processing and storage unit converts the event-related potential information into the event-related spectral disturbance information through Morlet wavelet, and converts the event-related spectral disturbance information into the functional connection information through phase delay index.

2. The auxiliary assessment system for autism as claimed in claim 1, wherein: During the training process of the evaluation model, multiple training data are input into multiple learning models, and from the output results of each learning model, a training data and a corresponding learning model with the highest accuracy are selected as the input data and the evaluation model respectively; the training data includes multiple typical cases and multiple atypical cases, and the multiple typical case data and multiple atypical case data are generated respectively after the question of the situational image, and the types of training data corresponding to the multiple typical case data and the multiple atypical case data include the first group of data, the second group of data and the third group of data.

3. The auxiliary assessment system for autism as claimed in claim 2, wherein: The classifier of the learning model includes at least two of a KNN classifier, a BDT classifier, a GAB classifier, an SVM classifier using a sigmoid kernel function, an SVM classifier using an RBF kernel function, and an SVM classifier using a polynomial kernel function.

4. The auxiliary assessment system for autism as claimed in claim 2, wherein: The training data comprises a plurality of sensors arranged at different locations according to the international 10-20 system of electroencephalography, wherein the locations include at least P3, P4, P7, and Pz, and each sensor corresponding to each location is defined as sensor P3, sensor P4, sensor P7, and sensor Pz; the event-related potential information in the first set of data in the training data includes information of sensor P7 within 160 to 220 ms after the event trigger point, and includes information of sensor Pz within 400 to 500 ms after the event trigger point; the event-related spectral disturbance information in the first set of data in the training data includes information of sensor P3 within 215 to 280 ms after the event trigger point and between a frequency of 3 and 5 Hz, and includes information of sensors P4 and Pz within 245 to 320 ms after the event trigger point and between a frequency of 4 and 6 Hz.

5. The auxiliary assessment system for autism as claimed in claim 2, wherein: The functional connection information in the second set of data in the training data includes information between 800 and 900 ms at a frequency of 4 to 7 Hz and information between 300 to 400 ms, 400 to 500 ms, and 800 to 900 ms at a frequency of 8 to 12 Hz.

6. The auxiliary assessment system for autism as claimed in claim 2, wherein: The training data is a plurality of sensors arranged at different positions according to the international 10-20 system of electroencephalography, wherein the positions include P3, P4, P7 and Pz, and each sensor corresponding to each position is defined as a sensor P3, a sensor P4, a sensor P7 and a sensor Pz respectively; the event-related potential information in the third set of data in the training data includes information of the sensor P7 160 to 220 ms after the event trigger point, and includes information of the sensor Pz 400 to 500 ms after the event trigger point; the event-related spectral disturbance information in the third set of data in the training data includes information of the sensor P3 215 to 280 ms after the event trigger point and between frequencies of 3 and 5 Hz, and includes information of the sensors P4 and Pz 245 to 320 ms after the event trigger point and between frequencies of 4 and 6 Hz; the functional connection information in the third set of data in the training data includes information of the sensor P7 160 to 220 ms after the event trigger point, and includes information of the sensor Pz 400 to 500 ms after the event trigger point; Hz in 800-900 ms, and information with a frequency between 8-12 Hz in 300-400 ms, 400-500 ms, and 800-900 ms.

7. The auxiliary assessment system for autism according to any one of claims 1 to 6, wherein: The classifier of the evaluation model is a SVM classifier using a sigmoid kernel function, and the input data is the first set of data.

Images