Listening test data validity evaluation method and system adopting artificial intelligence

Abstract

The invention relates to a hearing test data validity evaluation method and system adopting artificial intelligence, and relates to the technical field of big data analysis and evaluation, and the method comprises the steps: collecting hearing test data of a target object, and synchronously collecting a resting electroencephalogram signal of a preset resting period before each stimulus is given; acquiring historical hearing test data of a target object, and analyzing and acquiring a test evaluation input set in combination with the hearing test data and the resting electroencephalogram signal; and according to the test evaluation input set, activating a preset data evaluation channel to perform data validity evaluation, and obtaining a validity evaluation result. The problems that traditional hearing test data validity evaluation mostly depends on single test data, physiological state information of a test object and a historical test background are not fully associated, and the evaluation method is relatively simple and is easily influenced by environment or object state fluctuation, so that the evaluation accuracy is insufficient are solved.

Description

Technical Field

[0001] This application relates to the field of big data analysis and evaluation, and in particular to a method and system for evaluating the validity of hearing test data using artificial intelligence. Background Technology

[0002] With the increasing demands for test data quality in fields such as hearing health monitoring and rehabilitation effect evaluation, the reliability of hearing test data evaluation results has become a key technical requirement to support diagnostic and treatment decisions.

[0003] Currently, traditional methods for evaluating the validity of hearing test data do not fully integrate the physiological state information of the test subjects with the historical testing background. The evaluation methods often rely on a single data dimension, making it difficult to effectively eliminate the influence of environmental interference or fluctuations in the subject's state. This not only easily leads to biases in the judgment of validity, but also reduces the scientific rigor of subsequent diagnosis and treatment plans and rehabilitation effect assessments based on test data. Summary of the Invention

[0004] This application provides a method and system for evaluating the validity of hearing test data using artificial intelligence. It improves upon the problems of traditional hearing test data validity evaluation, which often relies on single test data, does not fully link the physiological state information of the test subject with the historical test background, and has a simple evaluation method that is easily affected by environmental or subject state fluctuations, thus leading to insufficient evaluation accuracy.

[0005] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide a method for evaluating the validity of hearing test data using artificial intelligence, the method comprising: Collect hearing test data of the target subjects and simultaneously collect resting EEG signals during a preset resting period before each stimulus is delivered; Historical hearing test data of the target object is obtained, and the hearing test data is combined with the resting EEG signal to analyze and obtain a test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability; Based on the test evaluation input set, activate the preset data evaluation channel to perform data validity evaluation and obtain the validity evaluation results.

[0006] Secondly, embodiments of this application provide a hearing test data validity evaluation system employing artificial intelligence, the system comprising: The hearing test and signal acquisition module is used to collect hearing test data of the target object and simultaneously acquire resting EEG signals during a preset resting period before each stimulus is delivered. The test evaluation input set acquisition module is used to acquire the historical hearing test data of the target object, and combine the hearing test data with the resting EEG signal to analyze and acquire the test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability; The data validity assessment module is used to activate a preset data assessment channel to perform data validity assessment based on the test assessment input set and obtain the validity assessment results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes an artificial intelligence-based method and system for evaluating the validity of hearing test data. Through a collaborative operation involving the simultaneous acquisition of hearing test data and resting EEG signals, the construction of a multi-dimensional test evaluation input set, the training of a data evaluation channel, and the execution of data validity evaluation, it achieves accurate judgment and reliable verification of the validity of hearing test data. First, by combining the target subject's historical hearing test data, historical resting repositioning time data is extracted to determine the hearing test interval. Simultaneously, the target subject's hearing test data and resting EEG signals from a preset resting period before each stimulus are acquired. Next, the target subject's historical hearing test data is obtained, and combined with the current hearing test data and resting EEG signals, resting-state fingerprint, test morphological stability, and noise co-occurrence stability are calculated, integrating them to form a test evaluation input set. Subsequently, multi-scale data reconstruction is performed based on the historical hearing test data to obtain a multi-scale sample dataset, and a deep learning-based data evaluation channel is trained and optimized. Finally, the preset data evaluation channel is activated according to the test evaluation input set, outputting a "valid" or "invalid" validity evaluation result as the core basis for judging the validity of the current hearing test data.

[0008] The technical solution of this application solves the problems of traditional hearing test data validity judgment relying on human experience, focusing only on test results while ignoring physiological state and environmental noise interference, low evaluation efficiency and poor consistency. It provides technical support for the reliable application of hearing test data and ensures that hearing health plans based on valid test data are scientific and suitable. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating a method for evaluating the validity of hearing test data using artificial intelligence, provided as an embodiment of this application; Figure 2 This is a schematic diagram of a hearing test data validity evaluation system using artificial intelligence, provided as an embodiment of this application.

[0011] The components represented by each number in the attached diagram are explained below: Hearing test and signal acquisition module 01, test evaluation input set acquisition module 02, data validity evaluation module 03. Detailed Implementation

[0012] This application provides a method and system for evaluating the validity of hearing test data using artificial intelligence, which addresses the technical problems in the prior art where the evaluation of the validity of hearing test data relies on a single data dimension, fails to fully integrate the physiological state information of the test subject and the historical test background, and has a simple evaluation logic that is easily affected by environmental interference or fluctuations in the state of the test subject, resulting in insufficient accuracy and reliability of the evaluation.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0015] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0016] Example 1, as shown in the appendix Figure 1As shown, this application provides a method for evaluating the validity of hearing test data using artificial intelligence, the method comprising the following steps: S110: Collect hearing test data of the target subject and simultaneously collect resting EEG signals during a preset resting period before each stimulus is delivered; In this embodiment of the application, in order to ensure that the collected hearing test data and resting EEG signals can accurately reflect the true hearing status of the target object, it is necessary to optimize the test interval by combining the target object's historical hearing test data, and then rely on the EEG sensing device to complete the synchronous data collection, so as to improve the accuracy and reliability of the subsequent effectiveness evaluation results.

[0017] Specifically, the historical resting time data is first extracted from the target subject's past hearing test data, which is related to the time required to recover to the resting state after each test.

[0018] The extraction of historical resting time data needs to cover test records of the target subject at different test stages and under different physical conditions to ensure that the data can fully reflect the true situation of resting time.

[0019] Furthermore, based on the extracted historical resting time data, the longest and least utilized time was selected and defined as the first interval of the hearing test, so as to ensure that the target subject has sufficient time to recover from the previous stimulus to the resting state.

[0020] Meanwhile, the fixed duration used to collect resting EEG signals is set as the second interval of the hearing test. The final hearing test interval is determined by calculating the sum of the first and second intervals.

[0021] After determining the hearing test interval and preset rest period, the target subject is given hearing stimulation sequentially according to the set test interval using an EEG sensing device. Resting EEG signals are collected during the preset rest period before each stimulation, and the hearing test data generated by the target subject under stimulation are recorded simultaneously.

[0022] This step, by combining historical hearing test data to optimize the test interval and simultaneously collecting two types of data, effectively avoids the problem of data being easily interfered with under the traditional fixed interval collection method, laying the foundation for accurately conducting the validity assessment of hearing test data.

[0023] Step S110 in the method provided in this application embodiment includes: Based on the target object's historical hearing test data, extract the historical resting repositioning time data; Based on the historical resting time data, the least useful time is selected to define the first interval of the hearing test; The preset resting period is used as the second interval of the hearing test, and the sum of the second interval of the hearing test and the first interval of the hearing test is calculated as the hearing test interval; Based on the hearing test interval and the preset resting period, the hearing test data and the resting EEG signal of the target object are collected synchronously by the EEG sensing device.

[0024] In this embodiment of the application, in order to ensure that the collected hearing test data and resting EEG signals can truly reflect the hearing and physiological state of the target object, it is necessary to optimize the test interval parameters by combining the target object's past test data, and then rely on professional sensing equipment to complete the synchronous collection of the two types of data, so as to improve the accuracy and reliability of the subsequent hearing test data validity assessment.

[0025] Specifically, the historical resting time data is first extracted based on the target subject's historical hearing test data.

[0026] Among them, historical hearing test data must cover records of the target subject at different testing stages and in different physical states. For example, it should include hearing test data in different scenarios such as healthy state, fatigue state, and mild ear discomfort state, so as to ensure that the extracted historical resting time data can fully reflect the actual time taken for the target subject to recover from the test stimulus to the resting state under various conditions.

[0027] During the extraction process, it is necessary to accurately locate the time node from "end of stimulation" to "return of EEG signal to resting baseline level" in each historical hearing test. The duration between the two time nodes is taken as the time for a single historical resting repositioning test. Then, all the single time data are integrated to form a complete set of historical resting repositioning time data.

[0028] After acquiring historical resting retrieval time data, the least efficient time was selected to define the first interval of the hearing test. The least efficient time is the longest duration in the historical resting retrieval time dataset. Selecting this duration as the first interval ensures that regardless of the subject's physical state, there is sufficient time to fully recover from the previous test stimulus to a resting state, avoiding the residual effects of previous stimuli from affecting the acquisition of subsequent resting EEG signals.

[0029] For example, if the shortest time to return to the target object in historical resting position data is 15 seconds, the longest time is 40 seconds, and the average time is 25 seconds, then 40 seconds is selected as the first interval for the hearing test.

[0030] Furthermore, a preset resting period is used as the second interval of the hearing test, and the sum of the second interval and the first interval of the hearing test is calculated as the hearing test interval.

[0031] The preset resting period is a fixed duration pre-set according to the requirements of EEG signal acquisition. This duration must meet the requirements for resting EEG signal feature extraction, for example, it is set to 20 seconds, to ensure that a sufficient length of resting EEG signal can be acquired for subsequent analysis.

[0032] Furthermore, by adding the first interval of the hearing test to the second interval of the hearing test, the final hearing test interval is obtained. For example, in the previous example, the first interval of 40 seconds is added to the second interval of 20 seconds to form a 60-second hearing test interval. This interval includes the recovery time of the target object and also reserves time for the acquisition of resting EEG signals.

[0033] Furthermore, based on the determined hearing test intervals and preset resting periods, the hearing test data and resting EEG signals of the target subjects are collected simultaneously using an EEG sensing device.

[0034] Specifically, the EEG sensing device should be a professional device with a high sampling rate and a high signal-to-noise ratio, such as an EEG machine with a sampling rate of not less than 250Hz, to ensure that the collected EEG signals can accurately reflect the activity state of brain neurons.

[0035] During the data collection process, the target subject is given auditory stimulation sequentially according to the set hearing test intervals. For example, pure tone signals of different frequencies and intensities are played through headphones as test stimuli. During the preset resting period before each stimulus is given, the EEG sensing device is activated to collect the target subject's resting EEG signals.

[0036] At the same time, the target's response data to stimuli is recorded through hearing testing equipment, such as whether the stimulus signal can be heard, the reaction time to the signal, and the ability to distinguish signals of different intensities, thus forming complete hearing test data.

[0037] During the synchronous acquisition process, it is necessary to ensure the time synchronization between devices. That is, the playback time of the auditory stimulus, the acquisition time of the resting EEG signal, and the recording time of the hearing test data must be precisely aligned with an error controlled within 10ms. This is to avoid data association errors caused by time deviations, thereby achieving accurate synchronous association between the hearing test data and the resting EEG signal, laying a data foundation for the subsequent construction of the test evaluation input set and the conduct of effectiveness evaluation.

[0038] S120: Obtain historical hearing test data of the target object, and combine the hearing test data with the resting EEG signal to analyze and obtain a test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability; In this embodiment of the application, in order to avoid the problem of biased evaluation and inability to eliminate accidental interference caused by relying solely on single test data, it is necessary to first obtain the target object's historical hearing test data, and then combine it with the current hearing test data and resting EEG signals for analysis, thereby obtaining a test evaluation input set containing multiple features, so as to achieve accurate judgment on the validity of the hearing test data.

[0039] Specifically, the first step is to obtain the target subject's historical hearing test data. Similarly, this historical data needs to cover records from different time periods and different testing scenarios for the target subject.

[0040] Furthermore, the acquired historical hearing test data is correlated and integrated with the currently collected hearing test data and resting EEG signals to lay the foundation for analyzing and obtaining the test evaluation input set.

[0041] During the integration process, it is necessary to ensure the consistency of data dimensions. For example, historical and current hearing test data should have the same test frequency range and the same response recording standards to avoid analysis bias caused by differences in data standards.

[0042] For example, if the target's historical hearing test used a frequency range of 125Hz-8000Hz, with "pressing a button within 1-3 seconds after hearing the signal" as a valid response, and the signal strength was adjusted in 10dBHL intervals, the current test must also match the frequency range of 125Hz-8000Hz, use the "pressing a button within 1-3 seconds after hearing the signal" response criterion, and maintain a 10dBHL signal strength adjustment interval to ensure that the historical and current data are completely consistent in the core test parameters.

[0043] At the same time, the resting EEG signals and the corresponding hearing test data need to be aligned along the time axis to ensure that each resting EEG signal can be accurately matched with its subsequent hearing test stimulus and response data, forming a complete data chain of "resting state - test stimulus - test response".

[0044] Furthermore, based on the integrated historical and current hearing test data, a test evaluation input set is obtained through analysis to construct a multi-dimensional basis for judging the validity of the hearing test data. This input set includes resting-state fingerprints, test morphological stability, and noise co-occurrence stability.

[0045] In the method provided in this application embodiment, "obtaining the resting fingerprint" includes: Historical resting EEG signals were extracted based on the historical hearing test data, and reference resting EEG signals were calculated using a fitting method. By comparing the reference resting EEG signal with the resting EEG signal, the EEG residual signal is obtained, and the multi-scale entropy of the EEG residual signal is calculated to obtain the first static fingerprint feature. Calculate the Lyapunov exponents of the reference resting EEG signal and the resting EEG signal to obtain the second static fingerprint feature; The first static fingerprint feature and the second static fingerprint feature are concatenated to obtain the resting fingerprint.

[0046] Specifically, the first step is to extract historical resting EEG signals based on historical hearing test data. During the extraction process, it is necessary to accurately locate the EEG signal segments corresponding to the "pre-set resting period before stimulus delivery" in each hearing test from the target subject's historical test records. These segments must meet the conditions of "no external interference and the target subject being in a natural relaxed state" to ensure that the extracted historical resting EEG signals can truly reflect the EEG characteristics of the target subject in a normal resting state.

[0047] At the same time, it is necessary to ensure that the number of extracted historical resting EEG signals is sufficient, such as covering at least 10 valid historical tests corresponding to different times and scenarios, so as to avoid deviation of subsequent reference signals due to insufficient sample size.

[0048] After acquiring historical resting EEG signals, a reference resting EEG signal was calculated using a fitting method.

[0049] Specifically, the fitting method should be an algorithm that can accurately fit the trend of EEG signals, such as the least squares method. By fitting the waveform characteristics, amplitude distribution, and frequency components of multiple historical resting EEG signals, an averaged waveform that can represent the EEG characteristics of the target object in a normal resting state is generated, i.e., the reference resting EEG signal.

[0050] For example, for each historical resting EEG signal, it is divided into several data windows along the time axis, the average amplitude in each window is calculated, and then the average amplitude of the windows corresponding to all historical signals is fitted to obtain the standard amplitude of the reference resting EEG signal in each time window. Finally, the signals are spliced ​​together to form a complete reference resting EEG signal, ensuring that the signal can comprehensively reflect the overall characteristics of the target object's historical resting EEG.

[0051] Furthermore, the reference resting EEG signal is compared with the resting EEG signal to obtain the EEG residual signal, and the multi-scale entropy of the EEG residual signal is calculated to obtain the first static fingerprint feature.

[0052] During the comparison process, the reference resting EEG signal and the currently acquired resting EEG signal need to be aligned along the time axis, and the amplitude difference between the two at the same time node is calculated point by point. The signal composed of these differences is the EEG residual signal. The magnitude of the EEG residual signal directly reflects the degree of deviation between the current resting EEG and the historical reference resting EEG signal.

[0053] Meanwhile, the calculation of multi-scale entropy requires setting a reasonable scale range, such as a scale interval of 1-10. By analyzing the complexity of the residual signal at different scales, the multi-scale entropy value is obtained as the first static fingerprint feature. This feature can reflect the irregularity of the current resting EEG signal. If the multi-scale entropy value is small in difference from the historical normal range of the target object, it indicates that the current resting state is highly consistent with the historical normal state.

[0054] Furthermore, the Lyapunov index of the reference resting EEG signal and the resting EEG signal is calculated to obtain the second static fingerprint feature.

[0055] Specifically, the Lyapunov index is used to measure the rate at which two initially similar signals separate over time. If the Lyapunov index of the reference resting EEG signal and the current resting EEG signal are small, it indicates that their chaotic characteristics are similar, and the current resting state is closer to the historical normal state. Conversely, if the Lyapunov index is large, it indicates that their chaotic characteristics are significantly different, and the current resting state may be abnormal.

[0056] During the calculation process, the phase space of the two types of signals needs to be reconstructed first to determine the appropriate embedding dimension and time delay. Then, the Lyapunov exponent is obtained by calculating the average separation rate of adjacent orbits, and this exponent is used as the second static fingerprint feature to reflect the consistency between the current resting state and the historical normal state from the perspective of chaotic characteristics.

[0057] Finally, the obtained first static fingerprint features and second static fingerprint features are concatenated to obtain the resting fingerprint.

[0058] During the splicing process, the multi-scale entropy value and the Lyapunov exponent need to be combined in a fixed order to form a two-dimensional feature vector. For example, the multi-scale entropy value can be used as the first dimension of the vector and the Lyapunov exponent as the second dimension, so as to ensure that the resting-state fingerprint constructed each time follows a unified feature structure.

[0059] The resting-state fingerprint obtained through the multi-dimensional feature splicing of the above steps can reflect the complexity of the current resting EEG signal and the degree of matching between the target object's current resting state and its historical routine state.

[0060] In the method provided in this application embodiment, "obtaining the stability of the test morphology" includes: Extract the historical sub-average waveform sequence from the historical hearing test data, and extract the current sub-average waveform sequence from the hearing test data; Calculate the dynamic time warping distance between the historical sub-average waveform sequence and the current sub-average waveform sequence to obtain the first morphological similarity index; Calculate the dynamic correlation coefficient of waveform shape between the historical sub-average waveform sequence and the current sub-average waveform sequence, and normalize the dynamic correlation coefficient to obtain the second morphological similarity index; The stability of the tested morphology is determined based on the weighted fusion result of the first morphological similarity index and the second morphological similarity index.

[0061] In this embodiment of the application, in order to avoid deviation in morphological similarity judgment due to a single waveform comparison method, it is necessary to extract historical and current waveform sequences, calculate multi-dimensional similarity indicators and perform weighted fusion to determine the stability of the test morphology, so as to provide a waveform-level basis for evaluating the validity of hearing test data.

[0062] Specifically, the historical sub-average waveform sequence is first extracted from the historical hearing test data, and the current sub-average waveform sequence is extracted from the hearing test data.

[0063] When extracting historical sub-average waveform sequences, the historical hearing test data should first be grouped according to the test frequency, such as by commonly used hearing test frequencies like 250Hz, 500Hz, 1000Hz, 2000Hz, 4000Hz, and 8000Hz, to ensure that each group covers the response waveforms of multiple historical tests at the same frequency.

[0064] Furthermore, the average amplitude of all historical test response waveforms within each group at each time point is calculated, and the average waveforms of each group are arranged in frequency order to form a historical sub-average waveform sequence.

[0065] When extracting the current sub-average waveform sequence, the frequency grouping standard must be completely consistent with that of the historical sub-average waveform sequence. For example, the groups should be divided according to the frequency range of 250Hz to 8000Hz. The average amplitude of the response waveform at each frequency in the current hearing test should be calculated and then arranged in the same frequency order to obtain the current sub-average waveform sequence, ensuring that the two are completely matched in terms of frequency dimension and data structure.

[0066] After obtaining the historical sub-average waveform sequence and the current sub-average waveform sequence, the dynamic time warping distance between the two is calculated to obtain the first morphological similarity index.

[0067] Specifically, the historical sub-average waveform sequence and the current sub-average waveform sequence are first preprocessed to remove baseline drift noise caused by minor equipment fluctuations or signal interference, ensuring that the amplitude changes of the waveform sequence can accurately reflect the true response characteristics of the target object to auditory stimuli.

[0068] Furthermore, based on the duration and amplitude variation range of the hearing test response waveform, time warping constraint parameters of the dynamic time warping algorithm are set, such as limiting the maximum ratio of time axis stretching or compression, to avoid excessive warping leading to waveform shape matching distortion.

[0069] Furthermore, the two preprocessed waveform sequences are input into the existing dynamic time warping algorithm. The algorithm automatically finds the time alignment path that minimizes the matching error between the two sequences and calculates the cumulative distance under the path. This cumulative distance is the dynamic time warping distance between the two. The smaller the distance value, the higher the morphological similarity between the historical and current sub-average waveform sequences, and vice versa. This distance value is used as the first morphological similarity index.

[0070] Furthermore, the dynamic correlation coefficient of waveform shape between the historical sub-average waveform sequence and the current sub-average waveform sequence is calculated, and the dynamic correlation coefficient is normalized to obtain the second morphological similarity index.

[0071] The waveform shape dynamic correlation coefficient measures the correlation between two waveform sequences in terms of amplitude variation trends. The closer the coefficient value is to 1, the more consistent the amplitude fluctuation trends and the higher the shape similarity between the two sequences. In specific calculations, a sliding window correlation analysis needs to be performed on each segment of the two waveform sequences to calculate the correlation coefficient between the two waveform segments within each window. Then, the correlation coefficients of all windows are integrated into a dynamic correlation coefficient.

[0072] Furthermore, the obtained dynamic correlation coefficients are normalized and mapped to a numerical range of 0-1 to eliminate the influence of differences in the absolute values ​​of the coefficients under different test scenarios, thus obtaining the second morphological similarity index.

[0073] Finally, the stability of the tested morphology is determined based on the weighted fusion result of the first morphological similarity index and the second morphological similarity index.

[0074] Before weighted fusion, reasonable weight coefficients need to be assigned based on the actual significance of the two indicators and the evaluation weight requirements. For example, since the dynamic time warp distance better reflects the matching degree of the overall waveform shape, it is given a weight of 0.6; the waveform shape dynamic correlation coefficient focuses more on the consistency of amplitude trend, and is given a weight of 0.4.

[0075] In the specific fusion calculation, the first morphological similarity index is first processed in reverse (because the smaller the distance value, the higher the similarity, the distance value needs to be converted into a value that is positively correlated with the similarity, that is, converted by the formula "1 / (1+dynamic time-normalized distance)"). Then, it is multiplied by the corresponding weights of the normalized second morphological similarity index and summed to obtain the weighted fusion result.

[0076] Finally, the stability of the test pattern is determined based on the magnitude of the weighted fusion result. For example, the closer the weighted fusion result is to 1, the higher the consistency between the historical and current waveform patterns, and the stronger the stability of the test pattern. Conversely, if the weighted fusion result is closer to 0, it indicates that the waveform patterns of the two are significantly different, and the stability of the test pattern is weaker.

[0077] Through the above steps, from waveform sequence extraction and multi-dimensional similarity calculation to weighted fusion, the differences in waveform morphology characteristics between historical and current hearing test responses were comprehensively analyzed, ensuring the accuracy of the determination results of test morphology stability and providing a waveform-level basis for subsequent evaluation of the validity of hearing test data.

[0078] In the method provided in this application embodiment, "obtaining the noise co-occurrence stability" includes: Obtain the historical hearing test data and the first and second test noise data of the hearing test data; Frequency domain analysis is performed on the first test noise data to identify multiple historical noise energy extreme frequency bands and generate the first noise frequency band distribution; Frequency domain analysis is performed on the second test noise data to identify multiple current noise energy extreme frequency bands and generate a second noise frequency band distribution; Using the energy proportions of corresponding frequency bands in the first and second noise frequency band distributions as weights, calculate the weighted residuals of the frequency center points between the first and second noise frequency band distributions; The noise co-occurrence stability is determined based on the statistical characteristics of the weighted residuals.

[0079] In this embodiment of the application, in order to determine the consistency of noise distribution between the current hearing test and the historical hearing test from the perspective of environmental noise, it is necessary to extract the noise data of the historical and current tests, analyze the noise frequency domain characteristics and calculate the weighted residuals, and then determine the noise co-occurrence stability, so as to provide a basis for environmental interference level for the validity assessment of hearing test data.

[0080] Specifically, the first step is to obtain historical hearing test data and the first and second test noise data of the hearing test data.

[0081] When acquiring the first test noise data, it is necessary to separate the background noise signal during the "period without hearing stimulation" from the historical hearing test data. These periods must correspond exactly to the periods of no signal output, such as the resting period and stimulation interval in the historical test, to ensure that the first test noise data can truly reflect the environmental noise situation during the historical test.

[0082] When acquiring the second test noise data, the separation criteria must be completely consistent with those of the first test noise data. The background noise signal of the same non-stimulation period should be extracted from the current hearing test data to ensure that the two are matched in terms of acquisition time and data length, so as to avoid subsequent analysis deviations due to differences in the noise extraction range.

[0083] Furthermore, frequency domain analysis is performed on the obtained first test noise data to identify multiple historical noise energy extreme frequency bands and generate the first noise frequency band distribution.

[0084] Specifically, frequency domain analysis requires the use of Fast Fourier Transform (a signal processing algorithm in the prior art) to convert the first test noise data in the time domain into frequency domain data, thereby obtaining the energy distribution of the noise signal at different frequencies.

[0085] Furthermore, an energy threshold is set, and frequency ranges with energy values ​​higher than the energy threshold are identified as noise energy extreme value frequency bands. For example, the energy threshold is set to 1.5 times the average noise energy in the entire frequency domain, and all frequency ranges with excessive energy are selected as historical noise energy extreme value frequency bands.

[0086] At the same time, for each historical noise energy extreme frequency band, its frequency range and the proportion of noise energy in that frequency band to the total noise energy are recorded. This information is organized in order from low to high frequency to generate the first noise frequency band distribution, which can clearly show the main energy concentration area of ​​historical test noise.

[0087] Furthermore, frequency domain analysis is performed on the second test noise data to identify multiple current noise energy extreme frequency bands and generate a second noise frequency band distribution.

[0088] Similarly, the analysis process must maintain completely consistent parameter settings with the frequency domain analysis of the first test noise data, that is, use the same fast Fourier transform parameters and the same energy threshold standard to ensure that the identification rules of the noise energy extreme frequency band are consistent.

[0089] Using the same steps, the current noise energy extreme value frequency band is selected from the frequency domain results of the second test noise data, and the frequency range and energy proportion of each frequency band are recorded. The second noise frequency band distribution is generated according to the same frequency sorting method, laying a unified foundation for subsequent comparative analysis with the first noise frequency band distribution.

[0090] Furthermore, using the energy proportions of corresponding frequency bands in the first and second noise frequency band distributions as weights, the weighted residuals of the frequency center points between the first and second noise frequency band distributions are calculated.

[0091] Specifically, the first step is to determine the corresponding frequency bands in the two types of noise frequency band distributions. That is, find the frequency bands in the first noise frequency band distribution and the second noise frequency band distribution where the frequency range overlaps by more than 80% and determine them as corresponding frequency bands. For non-overlapping frequency bands, if their energy proportion is less than 5% of the total noise energy, they are considered as secondary noise frequency bands and are not included in the calculation for the time being. If their energy proportion is higher than 5%, they are marked separately as abnormal frequency bands and will be given priority consideration later.

[0092] Simultaneously, for each corresponding frequency band, the frequency center points (i.e., the midpoints of the frequency ranges) of the first and second noise frequency bands are calculated, and the difference between the two center points is calculated as the original residual. At the same time, the average of the energy proportions of the corresponding frequency band in the first noise frequency band distribution and the energy proportions in the second noise frequency band distribution is taken as the weight of the corresponding frequency band. Finally, the original residual is multiplied by the corresponding weight to obtain the weighted residual of the corresponding frequency band. The weighted residuals of all corresponding frequency bands together constitute the weighted residual set.

[0093] Finally, the noise co-occurrence stability is determined based on the statistical characteristics of the weighted residuals. Specifically, the statistical characteristics of the weighted residual set are first calculated, such as the mean and variance of all weighted residuals.

[0094] The mean reflects the overall deviation of the center points of the corresponding frequency bands in the two types of noise frequency band distributions, while the variance reflects the dispersion of the deviations in each corresponding frequency band. Statistical characteristic thresholds are set, for example, the threshold for the weighted residual mean is set to 5Hz, and the variance threshold is set to 2Hz. 2 If the average weighted residual is less than 5Hz and the variance is less than 2Hz 2 If the noise coexistence stability is high, it indicates that the current test noise environment is consistent with the historical test noise environment, and the degree of noise interference on the test data is similar.

[0095] Conversely, if the average of the weighted residuals is greater than or equal to 5Hz or the variance is greater than or equal to 2Hz... 2 If the noise coexistence stability is low, it indicates that the current test noise environment is significantly different from the historical test noise environment, and there may be new strong interference noise sources. We need to be alert to the impact of noise on the accuracy of the current hearing test data.

[0096] The statistical characteristic-based judgment method described above can objectively reflect the stability of the noise environment and provide an effective environmental interference reference for evaluating the validity of hearing test data.

[0097] Finally, after obtaining the resting-state fingerprint, test morphological stability, and noise coexistence stability, these three types of features are integrated and serialized according to their feature structures to obtain the test evaluation input set. This ensures that the input set can fully cover the key information in the three dimensions of physiological state consistency, test waveform consistency, and noise environment consistency, providing a comprehensive analytical basis for activating the preset data evaluation channel to conduct a hearing test data validity evaluation.

[0098] S130: Based on the test evaluation input set, activate the preset data evaluation channel to perform data validity evaluation and obtain the validity evaluation result.

[0099] In this embodiment of the application, after completing the acquisition of hearing test data and resting EEG signals and constructing a test evaluation input set, in order to further determine whether the current hearing test data is reliable, it is necessary to activate the pre-constructed and trained data evaluation channel, and carry out automated evaluation based on the test evaluation input set in order to objectively obtain the validity evaluation results of the hearing test data.

[0100] Specifically, a data evaluation channel is first constructed by reconstructing, training, and optimizing models based on historical hearing test data at multiple scales, so as to have the ability to accurately evaluate the validity of hearing test data.

[0101] The method provided in this application embodiment includes the following steps for constructing the "data evaluation channel": Based on the historical hearing test data, multi-scale data reconstruction is performed to obtain a multi-scale sample dataset. A lightweight data evaluation channel based on deep learning is constructed, and the lightweight data evaluation channel is trained respectively using the multi-scale sample dataset and the first training iteration to obtain multiple alternative data evaluation channels. Acquire evaluation performance data from multiple alternative data evaluation channels, and combine the elbow method to determine the optimal data evaluation channel and the optimal scale sample data group; Multiple replica data evaluation channels are created for the optimal data evaluation channel. Random mutation channel configuration parameters are used to train multiple replica data evaluation channels according to the optimal scale sample data group and the second training number, wherein the first training number is less than the second training number. The data with the best evaluation performance is selected from the multiple replica data evaluation channels and the optimal data evaluation channel, and output as the data evaluation channel.

[0102] In this embodiment of the application, in order to construct an automated evaluation model that is both suitable for the scenario of evaluating the validity of hearing test data and has high accuracy, and to avoid the problems of traditional evaluation methods that rely on human experience, are inefficient and have large errors, it is necessary to construct a data evaluation channel through the steps of multi-scale sample construction, channel construction and multi-round training optimization, so as to provide model support for subsequent rapid and accurate judgment of data validity based on the test evaluation input set.

[0103] Specifically, firstly, multi-scale data reconstruction is performed based on historical hearing test data to obtain a multi-scale sample dataset, which covers sample information of different test scenarios and different data feature dimensions, providing a comprehensive training foundation for building a data evaluation channel.

[0104] The method provided in this application embodiment includes the following construction steps: "reconstructing multi-scale data based on the historical hearing test data to obtain a multi-scale sample dataset": Based on the historical hearing test data, N historical neighborhood data groups are established. Each historical neighborhood data group includes adjacent historical prior test data and historical subsequent test data, where N is a positive integer greater than 2. Traverse N historical neighborhood data groups, calculate the sample resting state fingerprint, sample test morphology stability and sample noise co-occurrence stability between the historical prior test data and the historical subsequent test data, and serialize and store the sample evaluation input set. The window length n is defined iteratively, and the sliding window method is used to perform sliding sorting on the sample evaluation input set to obtain a multi-scale sample evaluation input set, where n is a positive integer less than or equal to N; The multi-scale sample dataset is generated by matching the corresponding historical validity assessment results in the historical hearing test data according to the multi-scale sample assessment input set and storing them in association.

[0105] In this embodiment of the application, in order to avoid insufficient generalization ability of the model due to a single sample, it is necessary to complete the construction of a multi-scale sample dataset through the steps of historical data grouping, feature calculation, multi-scale sorting and result association, so as to ensure that the trained data evaluation channel can adapt to the needs of hearing test data validity evaluation in different scenarios.

[0106] Specifically, firstly, N historical neighborhood data groups (N is a positive integer greater than 2) are established based on historical hearing test data. When establishing historical neighborhood data groups, adjacent test data are paired according to the chronological order of the test time. For example, the test data from January 2024 (historical prior test data) and April 2024 (historical subsequent test data), and the test data from April 2024 and July 2024 are respectively formed into neighborhood data groups.

[0107] The value of N needs to be set reasonably based on the total amount of historical hearing test data. If the historical hearing test data contains 9 valid test records, then 8 historical neighborhood data groups (N=8) can be established to ensure that each group of data can reflect the characteristic changes within adjacent test periods.

[0108] Furthermore, after establishing N historical neighborhood data groups, the N historical neighborhood data groups are traversed to calculate the sample resting state fingerprint, sample test morphological stability, and sample noise co-occurrence stability between the historical prior test data and the historical subsequent test data, and the three types of parameters are serialized and stored to obtain the sample evaluation input set.

[0109] Specifically, when calculating the resting-state fingerprint of a sample, resting-state EEG signals need to be extracted from historical prior test data and historical subsequent test data respectively. Reference resting-state EEG signals are obtained through fitting methods, and the multi-scale entropy and Lyapunov exponent of the EEG residual signals of the two are calculated. The two are then spliced ​​together to form the resting-state fingerprint of the sample. When calculating the morphological stability of the sample test, the sub-average waveform sequences of the two sets of data are extracted, and the stability index is determined by the dynamic time-normalized distance and the normalized dynamic correlation coefficient. When calculating the noise co-occurrence stability of the sample, the test noise of the two sets of data is separated, and the stability result is obtained through frequency domain analysis and weighted residual calculation.

[0110] Furthermore, the three types of features corresponding to each group of historical neighborhood data are arranged in a fixed order of "sample resting state fingerprint - sample test morphological stability - sample noise co-occurrence stability" to form a feature sequence. The feature sequences of all groups are stored in the order in which the historical neighborhood data groups were established, and finally form a sample evaluation input set to ensure the orderliness and consistency of feature storage.

[0111] Furthermore, the window length n is defined iteratively, and the sliding window method is used to perform sliding sorting on the sample evaluation input set to obtain a multi-scale sample evaluation input set.

[0112] Specifically, the iteration range of the window length n is from 1 to N (n is a positive integer). For example, when N=8, n takes the values ​​1, 2, 3...8 in sequence. When n=1, the sliding window selects one feature sequence from the input set of the sample evaluation each time as a subset. When n=2, the sliding window selects two adjacent feature sequences each time to form a subset, and the window slides one feature sequence each time (step size is 1). For example, the first time it selects the 1st and 2nd features, the second time it selects the 2nd and 3rd features, until all feature sequences are covered.

[0113] Similarly, different n values ​​correspond to different scales of sample sets. Through the above iterative sliding method, the sample evaluation input set is decomposed into multiple sample evaluation input subsets of different scales. For example, when n=1, 8 subsets are obtained, and when n=2, 7 subsets are obtained. All subsets together constitute a multi-scale sample evaluation input set to achieve multi-dimensional partitioning of sample features.

[0114] Finally, based on the obtained multi-scale sample evaluation input set, the corresponding historical validity evaluation results are matched in the historical hearing test data and stored together to generate a multi-scale sample dataset.

[0115] Among them, the historical validity assessment results refer to the validity conclusions of the test data determined by manual review, professional equipment verification, etc. after the historical test is completed (including two types of assessment results: "valid" and "invalid"), which need to be extracted from the historical hearing test files.

[0116] During the matching process, it is necessary to find the historical validity assessment results corresponding to the historical prior and subsequent hearing test data in each sub-sample set of the multi-scale sample evaluation input set based on the historical neighborhood data group corresponding to each sub-sample set.

[0117] For example, if a subset contains historical neighborhood data groups 3 and 4, then the validity results of the pre-order (July 2024), post-order (October 2024) test data of group 3 and the pre-order (October 2024), post-order (January 2025) test data of group 4 need to be matched. Each subset is bound to all its corresponding historical validity evaluation results. For example, a subset with n=2 is associated with 4 historical validity results and stored in the structure of "subset features - historical validity evaluation results". All associated datasets are integrated to form a multi-scale sample dataset.

[0118] Furthermore, after acquiring multi-scale sample datasets, a data evaluation channel based on deep learning is constructed. Specifically, combining the feature dimensions (3 core features) for evaluating the validity of hearing test data with the real-time evaluation requirements, a lightweight convolutional neural network (MobileNet) is selected as the basic architecture. This architecture reduces the number of parameters and computational load through depthwise separable convolutions, which can improve operational efficiency while ensuring evaluation accuracy.

[0119] When building the architecture, the network parameters should be set reasonably according to the feature dimension and sample size of the multi-scale sample dataset: the input layer is set to a 3-dimensional feature vector (corresponding to resting fingerprint, test morphological stability, and noise co-occurrence stability), the hidden layer is set to 3-5 layers (including 2-3 convolutional layers and 1-2 pooling layers), the convolutional layer uses a 3×3 convolutional kernel (stride 1, same padding), the pooling layer uses 2×2 average pooling (stride 2), and the output layer is set to a 2-dimensional vector (corresponding to the two evaluation results of "valid" and "invalid").

[0120] Meanwhile, a batch normalization layer and a ReLU activation function are added after the convolutional layer to avoid gradient vanishing and accelerate training convergence. If the multi-scale sample dataset is small, the architecture can be further simplified by reducing one convolutional layer and pooling layer, and using a 1×1 convolutional kernel to reduce computational complexity; if the sample size is large, the number of neurons in the hidden layer can be appropriately increased to improve feature learning ability.

[0121] Furthermore, using the multi-scale sample dataset, the data evaluation channels are trained separately in the first training iteration to obtain multiple candidate data evaluation channels. Before training, the multi-scale sample dataset is divided into multiple subsets according to scale. For example, the samples corresponding to window lengths n=2, n=3, and n=4 are respectively used as three subsets.

[0122] For each subset of data, the same training parameters (learning rate 0.001, batch size 32) are used to train the data evaluation channels. The first training iteration is set to 50-80 rounds to ensure that the model initially converges on each subset. After training is completed for each subset, the channel model is saved once, resulting in multiple alternative data evaluation channels corresponding to the number of subsets. For example, 3 subsets correspond to 3 alternative channels, and each channel is adapted to sample data at a specific scale.

[0123] After obtaining multiple alternative data evaluation channels, their evaluation performance data are obtained, and the elbow method is used to determine the optimal data evaluation channel and the optimal scale sample data group.

[0124] The performance evaluation data mainly includes accuracy (the percentage of samples that correctly judge the validity of the data), recall (the percentage of valid data that are correctly identified), and F1 score (the harmonic mean of accuracy and recall). These three metrics are calculated for each candidate channel using the test set (divided from the multi-scale sample dataset, accounting for 20%).

[0125] Furthermore, a curve is plotted with the number of candidate channels on the horizontal axis and the mean F1 score of each channel on the vertical axis. The elbow method is used to find the elbow point of the curve, which is the inflection point where the mean F1 score increases from fast to slow as the number of channels increases.

[0126] For example, when the number of candidate channels increases from 1 to 2, the average F1 score increases from 85% to 91% (an increase of 6%); when it increases from 2 to 3, the average score increases from 91% to 92% (an increase of 1%). Then the number of channels corresponding to the elbow point is 2. At this time, the one with the higher F1 score among the two candidate channels is the optimal data evaluation channel, and the subset of data used for its training is the optimal scale sample data set.

[0127] Furthermore, multiple replica data evaluation channels are created based on the optimal data evaluation channel, the channel configuration parameters of each replica are randomly mutated, and the multiple replica data evaluation channels are trained according to the optimal scale sample data group, according to the second training iteration.

[0128] Specifically, firstly, the optimal data evaluation channel is copied to generate 5-10 replica channels. Then, the configuration parameters of each replica are randomly mutated: for example, adjusting the kernel size of the convolutional layer (from 3×3 to 5×5), modifying the learning rate (from 0.001 to 0.0005 or 0.002), changing the batch size (from 32 to 16 or 64), and increasing or decreasing the number of neurons in the hidden layer (±10%), to ensure that the parameters of each replica channel are different from the original channel but still within a reasonable range.

[0129] In addition, the second training iterations are set to 100-150 rounds (more than the first training iterations, to achieve model depth optimization). The optimal scale sample data set is used to train each replica channel. During the training process, the performance is monitored through the validation set (accounting for 20% of the optimal scale sample data set). Training is stopped when the F1 score on the validation set does not improve for 10 consecutive rounds, resulting in multiple trained replica data evaluation channels.

[0130] Finally, the channel with the best performance among multiple replica data evaluation channels and the optimal data evaluation channel is selected as the data evaluation channel. During the evaluation, the accuracy, recall, F1 score, and inference speed (single sample evaluation time) of all channels are calculated using the test set. After comprehensive ranking, the channel with the highest F1 score and inference speed ≤ 50ms / sample is selected as the optimal channel.

[0131] For example, if a replica channel has an F1 score of 94% and an inference speed of 42ms / sample, and the optimal original channel has an F1 score of 92% and an inference speed of 38ms / sample, while other replica channels all have F1 scores below 93%, then that replica channel is selected as the final data evaluation channel. If multiple channels have the same F1 score (e.g., all 94%), then the channel with the faster inference speed is selected to ensure a balance between accuracy and efficiency in the data evaluation channels.

[0132] Ultimately, the data evaluation channel constructed through the above steps can not only adapt to multi-scale hearing test feature inputs, but also has high-precision validity judgment capabilities and fast reasoning speed, providing stable and reliable technical support for the validity evaluation of hearing test data.

[0133] Furthermore, based on the test evaluation input set, a pre-built data evaluation channel is activated to perform data validity evaluation in order to obtain validity evaluation results.

[0134] Specifically, the test evaluation input set is first preprocessed and verified to check whether the features of the resting fingerprint are complete and whether the values ​​of test morphological stability and noise co-occurrence stability are within a reasonable range of 0-1, so as to avoid channel operation errors or result deviations due to abnormal input data.

[0135] Furthermore, after preprocessing, the pre-trained data evaluation channel is activated. The test evaluation input set is input into the data evaluation channel in a fixed order of "resting fingerprint - test morphological stability - noise co-occurrence stability". The data evaluation channel calls the feature weights and operation scheme learned during the training phase. First, it extracts the correlation information between the three types of features through the convolutional layer, then compresses redundant data through the pooling layer, and finally generates a "valid" or "invalid" validity evaluation result from the output layer.

[0136] For example, if the test evaluation input set for a target object includes: resting-state fingerprint (multi-scale entropy 0.75, Lyapunov exponent 0.21), test morphological stability 0.88, and noise co-occurrence stability 0.92, and preprocessing verification confirms that all features are complete and their values ​​are all in the 0-1 range, after inputting them into the data evaluation channel in a fixed order, the channel identifies the associated features that "the resting-state fingerprint conforms to the historical normal range, and the test morphology and noise environment stability are both high" through the convolutional layer. After compression by the pooling layer, the output layer generates a "valid" validity evaluation result, which intuitively reflects the reliability of the hearing test data.

[0137] At the same time, the results of this effectiveness evaluation should be associated with and stored with the corresponding test evaluation input set and channel operation logs, and archived according to the target object ID and test time. This will provide real-world case support for the iterative optimization of the data evaluation channel and further improve the evaluation accuracy of the channel in different test scenarios.

[0138] Ultimately, the effectiveness assessment results obtained through the above steps will serve as the core basis for determining whether the current hearing test data can be used for subsequent diagnosis and treatment or hearing status analysis, so as to ensure that the hearing health plan based on reliable data is scientific and appropriate, thereby ensuring the accuracy and effectiveness of hearing health management.

[0139] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes an artificial intelligence-based method for evaluating the validity of hearing test data. First, it combines historical hearing test data of the target subject to extract historical resting time data and selects the least efficient time to define the first interval of the hearing test. A preset resting period is used as the second interval to determine the final test interval. Hearing test data and resting EEG signals from the preset resting period before each stimulus are simultaneously acquired using an EEG sensor. Next, historical hearing test data of the target subject is obtained, and combined with current hearing test data and resting EEG signals, resting-state fingerprint, test morphological stability, and noise co-occurrence stability are calculated and integrated to form a test evaluation input set. Subsequently, a multi-scale sample dataset is constructed based on historical hearing test data, and a deep learning-based data evaluation channel is trained and optimized. Finally, the test evaluation input set is preprocessed and validated, and the activated data evaluation channel is input in a fixed order to generate a "valid" or "invalid" validity evaluation result. Evaluation records are associated and stored to support iterative channel optimization, thereby achieving intelligent judgment of the validity of hearing test data.

[0140] The method provided in this application solves the problems of traditional hearing test data validity judgment relying on human experience, focusing only on test results while ignoring physiological state and environmental noise interference, and having low evaluation efficiency and poor consistency by adopting the technical solution of "optimizing test interval with historical data - synchronously collecting test data and EEG signals - constructing test evaluation input set with multi-dimensional features - training data evaluation channel with multi-scale sample data - outputting validity evaluation results". It realizes intelligent evaluation of the entire process from data collection to result output, and provides technical support for the effective application of hearing test data.

[0141] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of a hearing test data validity evaluation method using artificial intelligence provided in Embodiment 1, this application also provides a hearing test data validity evaluation system using artificial intelligence, specifically including: The hearing test and signal acquisition module 01 is used to acquire the hearing test data of the target object and simultaneously acquire the resting EEG signal of a preset resting period before each stimulus is given. The test evaluation input set acquisition module 02 is used to acquire the historical hearing test data of the target object, and combine the hearing test data with the resting EEG signal to analyze and acquire the test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability. The data validity assessment module 03 is used to activate a preset data assessment channel to perform data validity assessment based on the test assessment input set and obtain the validity assessment result.

[0142] In one embodiment, the hearing test and signal acquisition module 01 is further used for: Based on the target object's historical hearing test data, historical resting repositioning time data is extracted; according to the historical resting repositioning time data, the least utilized time is selected to define the first hearing test interval; the preset resting period is used as the second hearing test interval, and the sum of the second hearing test interval and the first hearing test interval is calculated as the hearing test interval; according to the hearing test interval and the preset resting period, the target object's hearing test data and resting EEG signals are synchronously collected using an EEG sensing device.

[0143] In one embodiment, the test evaluation input set acquisition module 02 further includes: Historical resting EEG signals are extracted based on the historical hearing test data, and a reference resting EEG signal is calculated using a fitting method. The reference resting EEG signal is compared with the resting EEG signal to obtain the EEG residual signal, and the multi-scale entropy of the EEG residual signal is calculated to obtain the first static fingerprint feature. The Lyapunov exponent of the reference resting EEG signal and the resting EEG signal is calculated to obtain the second static fingerprint feature. The first static fingerprint feature and the second static fingerprint feature are concatenated to obtain the resting-state fingerprint.

[0144] Furthermore, the test evaluation input set acquisition module 02 also includes: Historical sub-average waveform sequences are extracted from the historical hearing test data, and current sub-average waveform sequences are extracted from the hearing test data; the dynamic time warping distance between the historical sub-average waveform sequences and the current sub-average waveform sequences is calculated to obtain a first morphological similarity index; the waveform shape dynamic correlation coefficient between the historical sub-average waveform sequences and the current sub-average waveform sequences is calculated, and the dynamic correlation coefficient is normalized to obtain a second morphological similarity index; the stability of the test morphology is determined based on the weighted fusion result of the first morphological similarity index and the second morphological similarity index.

[0145] Furthermore, the test evaluation input set acquisition module 02 also includes: The process involves acquiring historical hearing test data, first test noise data, and second test noise data; performing frequency domain analysis on the first test noise data to identify multiple historical noise energy extreme frequency bands and generating a first noise frequency band distribution; performing frequency domain analysis on the second test noise data to identify multiple current noise energy extreme frequency bands and generating a second noise frequency band distribution; calculating the weighted residual between the frequency center points of the first and second noise frequency band distributions using the energy proportion of corresponding frequency bands in the first and second noise frequency band distributions as weights; and determining the noise co-occurrence stability based on the statistical characteristics of the weighted residual.

[0146] In one embodiment, the data validity assessment module 03 further includes: Based on the historical hearing test data, multi-scale data reconstruction is performed to obtain a multi-scale sample dataset; a lightweight data evaluation channel based on deep learning is constructed, and the lightweight data evaluation channel is trained using the multi-scale sample dataset and a first training iteration to obtain multiple candidate data evaluation channels; the evaluation performance data of the multiple candidate data evaluation channels are obtained, and the optimal data evaluation channel and the optimal scale sample data group are determined using the elbow method; multiple replica data evaluation channels of the optimal data evaluation channel are created, the channel configuration parameters are randomly mutated, and the multiple replica data evaluation channels are trained according to the optimal scale sample data group and a second training iteration, wherein the first training iteration is less than the second training iteration; the data with the best evaluation performance data is selected from the multiple replica data evaluation channels and the optimal data evaluation channel, and output as the data evaluation channel.

[0147] Furthermore, the data validity assessment module 03 also includes: Based on the historical hearing test data, N historical neighborhood data groups are established, each including adjacent historical prior test data and historical subsequent test data, where N is a positive integer greater than 2. The N historical neighborhood data groups are traversed, and the resting-state fingerprint, morphological stability, and noise co-occurrence stability of the samples between the historical prior test data and the historical subsequent test data are calculated respectively. The samples are then serialized and stored to obtain a sample evaluation input set. A window length n is iteratively defined, and the sample evaluation input set is sorted using a sliding window method to obtain a multi-scale sample evaluation input set, where n is a positive integer less than or equal to N. The multi-scale sample evaluation input set is matched with the corresponding historical validity evaluation results in the historical hearing test data, and stored in association to generate the multi-scale sample dataset.

[0148] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0149] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0150] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for evaluating the validity of hearing test data using artificial intelligence, characterized in that, include: Collect hearing test data of the target subjects and simultaneously collect resting EEG signals during a preset resting period before each stimulus is delivered; Historical hearing test data of the target object is obtained, and the hearing test data is combined with the resting EEG signal to analyze and obtain a test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability; Based on the test evaluation input set, activate the preset data evaluation channel to perform data validity evaluation and obtain the validity evaluation results.

2. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 1, characterized in that, Hearing test data of the target subjects were collected, and resting EEG signals during a pre-defined resting period before each stimulus was delivered were simultaneously collected, including: Based on the target object's historical hearing test data, extract the historical resting repositioning time data; Based on the historical resting time data, the least useful time is selected to define the first interval of the hearing test; The preset resting period is used as the second interval of the hearing test, and the sum of the second interval of the hearing test and the first interval of the hearing test is calculated as the hearing test interval; Based on the hearing test interval and the preset resting period, the hearing test data and the resting EEG signal of the target object are collected synchronously by the EEG sensing device.

3. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 2, characterized in that, Obtaining the resting fingerprint includes: Historical resting EEG signals were extracted based on the historical hearing test data, and reference resting EEG signals were calculated using a fitting method. By comparing the reference resting EEG signal with the resting EEG signal, the EEG residual signal is obtained, and the multi-scale entropy of the EEG residual signal is calculated to obtain the first static fingerprint feature. Calculate the Lyapunov exponents of the reference resting EEG signal and the resting EEG signal to obtain the second static fingerprint feature; The first static fingerprint feature and the second static fingerprint feature are concatenated to obtain the resting fingerprint.

4. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 2, characterized in that, Obtaining the stability of the test morphology includes: Extract the historical sub-average waveform sequence from the historical hearing test data, and extract the current sub-average waveform sequence from the hearing test data; Calculate the dynamic time warping distance between the historical sub-average waveform sequence and the current sub-average waveform sequence to obtain the first morphological similarity index; Calculate the dynamic correlation coefficient of waveform shape between the historical sub-average waveform sequence and the current sub-average waveform sequence, and normalize the dynamic correlation coefficient to obtain the second morphological similarity index; The stability of the tested morphology is determined based on the weighted fusion result of the first morphological similarity index and the second morphological similarity index.

5. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 2, characterized in that, Obtaining the noise co-occurrence stability includes: Obtain the historical hearing test data and the first and second test noise data of the hearing test data; Frequency domain analysis is performed on the first test noise data to identify multiple historical noise energy extreme frequency bands and generate the first noise frequency band distribution; Frequency domain analysis is performed on the second test noise data to identify multiple current noise energy extreme frequency bands and generate a second noise frequency band distribution; Using the energy proportions of corresponding frequency bands in the first and second noise frequency band distributions as weights, calculate the weighted residuals of the frequency center points between the first and second noise frequency band distributions; The noise co-occurrence stability is determined based on the statistical characteristics of the weighted residuals.

6. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 1, characterized in that, The construction steps of the data evaluation channel include: Based on the historical hearing test data, multi-scale data reconstruction is performed to obtain a multi-scale sample dataset. A lightweight data evaluation channel based on deep learning is constructed, and the lightweight data evaluation channel is trained respectively using the multi-scale sample dataset and the first training iteration to obtain multiple alternative data evaluation channels. Acquire evaluation performance data from multiple alternative data evaluation channels, and combine the elbow method to determine the optimal data evaluation channel and the optimal scale sample data group; Multiple replica data evaluation channels are created for the optimal data evaluation channel. Random mutation channel configuration parameters are used to train multiple replica data evaluation channels according to the optimal scale sample data group and the second training number, wherein the first training number is less than the second training number. The data with the best evaluation performance is selected from the multiple replica data evaluation channels and the optimal data evaluation channel, and output as the data evaluation channel.

7. The method for evaluating the validity of hearing test data using artificial intelligence as described in claim 6, characterized in that, Based on the historical hearing test data, multi-scale data reconstruction is performed to obtain a multi-scale sample dataset, including: Based on the historical hearing test data, N historical neighborhood data groups are established. Each historical neighborhood data group includes adjacent historical prior test data and historical subsequent test data, where N is a positive integer greater than 2. Traverse N historical neighborhood data groups, calculate the sample resting state fingerprint, sample test morphology stability and sample noise co-occurrence stability between the historical prior test data and the historical subsequent test data, and serialize and store the sample evaluation input set. The window length n is defined iteratively, and the sliding window method is used to perform sliding sorting on the sample evaluation input set to obtain a multi-scale sample evaluation input set, where n is a positive integer less than or equal to N; The multi-scale sample dataset is generated by matching the corresponding historical validity assessment results in the historical hearing test data according to the multi-scale sample assessment input set and storing them in association.

8. A hearing test data validity evaluation system employing artificial intelligence, characterized in that, The system is used to execute the method for evaluating the validity of hearing test data using artificial intelligence as described in any one of claims 1-7, the system comprising: The hearing test and signal acquisition module is used to collect hearing test data of the target object and simultaneously acquire resting EEG signals during a preset resting period before each stimulus is delivered. The test evaluation input set acquisition module is used to acquire the historical hearing test data of the target object, and combine the hearing test data with the resting EEG signal to analyze and acquire the test evaluation input set, wherein the test evaluation input set includes at least resting fingerprint, test morphological stability and noise coexistence stability; The data validity assessment module is used to activate a preset data assessment channel to perform data validity assessment based on the test assessment input set and obtain the validity assessment results.

Images