Intelligent auditory sense screening system based on machine learning

Abstract

The invention relates to the technical field of auditory screening, and discloses a machine learning-based intelligent auditory screening system, which comprises an auditory measurement end and a background server, the hearing measurement end is used for detecting hearing data of a patient / user in a measurement period and storing the hearing data to the background server; the background server comprises a data module which is used for storing real-time hearing data and historical hearing state data of a patient / user. According to the intelligent hearing screening system based on machine learning, a convenient hearing test mode is provided for common users and doctors by adopting a big data technology and designing the convenient hearing test mode and hearing analysis based on big data, and rapid and accurate hearing test experience is achieved; meanwhile, according to big data analysis, the hearing loss trend is analyzed according to historical hearing test records, proper suggestions are provided for the user and doctors, and the user is helped to correctly understand the hearing condition of himself / herself.

Description

Technical Field

[0001] This invention relates to the field of hearing screening technology, specifically to an intelligent hearing screening system based on machine learning. Background Technology

[0002] The most basic and important hearing test method in clinical practice is the pure-tone audiometry test. It is one of the subjective behavioral test methods that can accurately reflect hearing sensitivity. Its purpose is to reflect the hearing level of the smallest sound at each frequency that the subject can hear in a quiet environment. The accuracy rate of its test results can reach 75%-80%, and it is recognized as the "gold standard" for assessing hearing. When performing pure-tone audiometry in clinical practice, the Hughson-Westlake method, also known as the 5-point rise and 10-point fall method, is often used. It is an examination based on a certain test sequence and steps, and is particularly suitable for automated mode.

[0003] While existing pure-tone audiometry can screen hearing, its use is limited in several aspects, primarily due to the following shortcomings:

[0004] 1) The high requirements of pure tone hearing threshold testing for sound insulation and noise reduction environment have limited the popularization of hearing tests. Although some software or hardware for automatic pure tone hearing threshold testing have reduced the high requirements for the environment through various means, the accuracy of sound frequency and intensity and the reliability of the results have been reduced.

[0005] 2) The duration of the pure tone audiometry test cannot be well controlled. The existing pure tone audiometry tests all start with the same fixed frequency and sound intensity. However, in the actual screening process, the hearing level of the subjects varies, and subjects with hearing loss will take more time. In addition, subjects of different ages have different comprehension and operational abilities, which will also affect the test duration.

[0006] 3) The data storage mode of the pure tone hearing threshold automatic test is relatively simple, and it mostly adopts the mode of uniformly uploading to the backend database for storage. This mode is not conducive to quickly obtaining relevant information of the subjects and is not conducive to further diagnosis and treatment intervention in the later stage.

[0007] 4) When ordinary users use some software for testing, they can only get their own test results. They do not have a clear understanding of their own hearing level or the trend of hearing loss decline or increase. This is not conducive to users understanding their hearing status and taking timely intervention measures, thus missing the best time for diagnosis and treatment.

[0008] Therefore, a machine learning-based intelligent hearing screening system is proposed to solve the problems mentioned above. Summary of the Invention

[0009] (a) Technical problems to be solved

[0010] To address the shortcomings of existing technologies, this invention provides an intelligent hearing screening system based on machine learning. It utilizes big data technology, designs convenient hearing testing methods, and performs big data-based hearing analysis, offering a convenient hearing testing experience for both ordinary users and doctors, achieving a fast and accurate hearing test experience. Furthermore, based on big data analysis, it analyzes historical hearing test records to identify hearing loss trends, providing appropriate suggestions to users and doctors, helping users correctly understand their hearing status. This invention solves the following problems: 1) Pure-tone audiometry requires a high level of noise reduction and sound insulation, limiting its widespread adoption. Although some pure-tone audiometry software or hardware has reduced these environmental requirements through various means, the accuracy of sound frequency and intensity, as well as the reliability of the results, has decreased; 2) The duration of pure-tone audiometry testing cannot be well controlled. Some automated pure-tone audiometry tests start with the same fixed frequency and intensity. However, in actual screening, the varying hearing levels of the subjects result in longer testing times for those with hearing loss. Furthermore, differences in comprehension and operational abilities among subjects of different ages also affect the test duration. 3) Automated pure-tone audiometry tests use a relatively simple data storage model, often uploading data to a backend database. This model is not conducive to quickly obtaining relevant information about the subjects or to further diagnostic and treatment interventions. 4) When ordinary users use certain software for testing, they only receive their own test results and lack a clear understanding of their hearing level or the trend of hearing loss, hindering their ability to monitor their hearing status and take timely interventions, potentially leading to missed opportunities for optimal treatment.

[0011] (II) Technical Solution

[0012] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: an intelligent hearing screening system based on machine learning, including a hearing measurement terminal and a back-end server;

[0013] The hearing test terminal: detects patient / user hearing data within the test period and saves it to the backend server;

[0014] The backend server includes:

[0015] Data module: Stores real-time hearing data and historical hearing status data for patients / users;

[0016] Analysis controller: Based on preset hearing data and hearing classification comparison strategy, determine the hearing status of the patient / user based on real-time saved hearing data;

[0017] Hearing trend module: Based on the analysis controller, the system determines the patient / user's hearing status in real time and analyzes the patient / user's historical hearing status data to generate hearing trend results;

[0018] Hearing test results module: Based on preset hearing data and hearing classification comparison strategies, the analysis controller determines the patient / user's hearing status and hearing trend in real time. The module generates hearing trend results data for patients / users as the basis for judgment, and generates hearing suggestions and trend text.

[0019] The beneficial effects of this invention are:

[0020] This machine learning-based intelligent hearing screening system utilizes big data technology to provide ordinary users and doctors with a convenient hearing test method and big data-based hearing analysis, enabling a fast and accurate hearing test experience. Furthermore, based on big data analysis, it analyzes historical hearing test records to identify hearing loss trends and provides appropriate suggestions to users and doctors, helping users correctly understand their hearing condition.

[0021] Based on the above technical solution, the present invention can be further improved as follows.

[0022] Furthermore, the hearing measurement terminal includes:

[0023] Ambient noise detection module: Detects ambient noise data of the patient / user's environment during the measurement period;

[0024] Hearing test module: Separates environmental noise data and tests patient / user hearing data within the measurement period.

[0025] Furthermore, the hearing classification in the preset hearing data and hearing classification comparison strategy of the analysis controller is as follows:

[0026] Level 1: The user's or patient's life is basically unaffected;

[0027] Level 2: The user's or patient's life will be affected, and communication will be hindered;

[0028] Level 3: The user or patient's life is severely affected, and communication is extremely difficult;

[0029] Level 4: The user or patient has no problem with hearing communication, but has difficulty locating the sound source.

[0030] Furthermore, the hearing trend module generates hearing trend results based on the real-time determination of the patient / user's hearing status by the analysis controller and the analysis of the patient / user's historical status data, specifically as follows:

[0031] Predicting hearing trends using time series models;

[0032] The time series model function is as follows:

[0033] f(x) = X1 + k(t) n -t1)

[0034] Where f(x) is a time series, X1, X2, ..., X... n The hearing level is graded for each test, k represents the hearing trend, and t1, t2, ..., t3 are the gradations for each test. n For time;

[0035] Where t1 corresponds to X1, t2 corresponds to X2, ..., tt n Corresponding to X n .

[0036] Furthermore, the hearing trend module generates hearing trend results based on the real-time determination of the patient / user's hearing status by the analysis controller and the analysis of the patient / user's historical hearing status data, specifically as follows:

[0037] S1. Construct a Gaussian process model;

[0038] S2, Based on dataset D = {(t n ,X n Training a Gaussian process model;

[0039] Among them, t n It is the time point for the hearing test, and X n This corresponds to the hearing level;

[0040] S3. Predicts real-time patient / user hearing trend results based on trained Gaussian process models;

[0041] S4. Output the predicted hearing trend results.

[0042] Furthermore, the Gaussian process model function is as follows:

[0043]

[0044] Where m(t) is the mean function describing the central tendency of the data points, and k(s,t) is the kernel function defining the correlation between time points s and t, where s and t represent any two time points.

[0045] Furthermore, the kernel function k(s,t) is a radial basis function (RBF), as follows:

[0046]

[0047] Where, σ 2 It is the variance, where s and t are two distinct time points in the continuous domain of the Gaussian process domain, ||st||. 2It is the distance between two different time points s and t.

[0048] Furthermore, the hearing test result module, based on preset hearing data and hearing classification comparison strategies, analyzes and controls the real-time determination of the patient / user's hearing status and hearing trend module to generate patient / user hearing trend result data as the basis for judgment, generating hearing suggestions and trend text, specifically:

[0049] S1. Construct a hidden Markov model based on hearing data;

[0050] S2. Train the Hidden Markov Model;

[0051] S3. Generate real-time patient / user hearing suggestions and trend text based on trained Hidden Markov Models;

[0052] S4. Output listening suggestions and trend text.

[0053] Furthermore, initialization: select an initial state based on the initial state probability.

[0054] Generate an observation sequence: Generate an observation based on the current state and the observation probability.

[0055] State transition: The state transitions to the next state based on the current state and the state transition probability.

[0056] Repeated generation and transfer: Repeat the above steps until a predetermined termination condition is met, such as generating a specific text length or encountering a terminator.

[0057] Generate complete sentences of listening trend conclusions and recommendations using a hidden Markov model. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the structure of the present invention;

[0059] Figure 2 This is a flowchart illustrating the structure of the present invention.

[0060] In the diagram: 1. Hearing test terminal; 2. Backend server; 3. Data module; 4. Analysis controller; 5. Hearing trend module; 6. Hearing test result module; 7. Ambient sound detection module; 8. Hearing test module. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] In the embodiments, by Figure 1-2 The present invention provides an intelligent hearing screening system based on machine learning, comprising a hearing measurement terminal 1 and a back-end server 2.

[0063] Hearing test terminal 1: Detects patient / user hearing data within the testing period and saves it to the backend server terminal 2;

[0064] Backend server 2 includes:

[0065] Data Module 3: Saves real-time hearing data and historical hearing status data of patients / users;

[0066] Analysis Controller 4: Based on preset hearing data and hearing classification comparison strategies, determine the patient / user's hearing status based on real-time saved hearing data;

[0067] Hearing trend module 5: Based on the analysis controller 4, the system determines the patient / user's hearing status in real time and analyzes the patient / user's historical hearing status data to generate hearing trend results;

[0068] Hearing test results module 6: Based on preset hearing data and hearing classification comparison strategy, the analysis controller 4 determines the patient / user's hearing status and hearing trend module 5 in real time, and generates hearing trend results data for the patient / user as the basis for judgment to generate hearing suggestions and trend text.

[0069] The backend server 2 can be divided into a regular user client and a doctor user client.

[0070] Among them, the ordinary user terminal is for ordinary users who need to take a hearing test. It provides an individual hearing test port for ordinary users and transmits the relevant data after the test to the data module 3 for storage.

[0071] Doctor User App: Provides management of information for ordinary users, management of hearing test information for ordinary users, and transfers the corresponding data to data module 3 for storage.

[0072] Among them, the ordinary user terminal includes a smart terminal installed on the ordinary user's mobile device, which provides user information management and hearing test functions;

[0073] The doctor's user interface includes smart terminal devices used by hospitals to manage and analyze patients' hearing status. These smart terminal devices manage and analyze patient information and hearing test data, providing convenient management functions for medical staff.

[0074] Regular users can choose between the standard version and the professional version for testing.

[0075] The standard version tests key items, has fewer test items, and takes less time, making it suitable for quickly assessing a user's hearing status.

[0076] The professional version offers comprehensive testing with numerous items, taking a considerable amount of time, making it suitable for a complete assessment of a user's hearing condition.

[0077] The hearing measurement terminal 1 includes:

[0078] Ambient noise detection module 7: Detects ambient noise data of the patient / user's environment during the measurement period;

[0079] Hearing test module 8: Separates environmental noise data and tests patient / user hearing data within the measurement period.

[0080] Among them, the hearing classification in the preset hearing data and hearing classification comparison strategy of the analysis controller 4 is as follows:

[0081] Level 1: The user's or patient's life is basically unaffected;

[0082] Level 2: The user's or patient's life will be affected, and communication will be hindered;

[0083] Level 3: The user or patient's life is severely affected, and communication is extremely difficult;

[0084] Level 4: The user or patient has no problem with hearing communication, but has difficulty locating the sound source.

[0085] Among them, the hearing trend module 5 generates hearing trend results based on the real-time determination of the patient / user's hearing status and the patient / user's historical status data by the analysis controller 4. Specifically:

[0086] Predicting hearing trends using time series models;

[0087] The time series model function is as follows:

[0088] f(x) = X1 + k(t) n -t1)

[0089] Where f(x) is a time series, X1, X2, ..., X... n The hearing level is graded for each test, k represents the hearing trend, and t1, t2, ..., t3 are the gradations for each test. n For time;

[0090] Where t1 corresponds to X1, t2 corresponds to X2, ..., tt n Corresponding to X n .

[0091] Among them, the hearing trend module 5 generates hearing trend results based on the real-time determination of the patient / user's hearing status and the patient / user's historical hearing status data by the analysis controller 4. Specifically:

[0092] S1. Construct a Gaussian process model;

[0093] S2, Based on dataset D = {(t n ,X n Training a Gaussian process model;

[0094] Among them, t n It is the time point for the hearing test, and X n This corresponds to the hearing level;

[0095] S3. Predicts real-time patient / user hearing trend results based on trained Gaussian process models;

[0096] S4. Output the predicted hearing trend results.

[0097] The Gaussian process model function is as follows:

[0098]

[0099] Where m(t) is the mean function describing the central tendency of the data points, and k(s,t) is the kernel function defining the correlation between time points s and t, where s and t represent any two time points.

[0100] Wherein, the kernel function k(s,t) is the radial basis function (RBF), as follows:

[0101]

[0102] Where, σ 2 It is the variance, where s and t are two distinct time points in the continuous domain of the Gaussian process domain, ||st||. 2 It is the distance between two different time points s and t.

[0103] The Gaussian function GP(m(t),k(t,s)) is determined based on:

[0104] Since s and t are two distinct time points within the continuous domain of a Gaussian process, i.e., at the times of two different hearing tests, ||st|| 2 The distance between two different time points s and t is the distance between them. The radial basis function outputs a scalar, which represents the covariance between the two different hearing test time points in the Gaussian distribution. The greater the distance between the two time points, the smaller the covariance between the two distributions. In other words, the greater the time interval, the weaker the correlation between the two hearing tests. Conversely, the closer the two time points are, the greater the covariance between the distributions, and the stronger the correlation between the two hearing tests. Thus, the Gaussian function is determined.

[0105] Among them, the hearing test result module 6 determines the patient / user's hearing status and hearing trend in real time based on preset hearing data and hearing classification comparison strategy, and the analysis controller 4 generates patient / user hearing trend result data as the basis for judgment to generate hearing suggestions and trend text, specifically:

[0106] S1. Construct a hidden Markov model based on hearing data;

[0107] S2. Train the Hidden Markov Model;

[0108] S3. Generate real-time patient / user hearing suggestions and trend text based on trained Hidden Markov Models;

[0109] S4. Output listening suggestions and trend text.

[0110] Initialization: Select an initial state based on the initial state probability.

[0111] Generate an observation sequence: Generate an observation based on the current state and the observation probability.

[0112] State transition: The state transitions to the next state based on the current state and the state transition probability.

[0113] Repeated generation and transfer: Repeat the above steps until a predetermined termination condition is met, such as generating a specific text length or encountering a terminator.

[0114] Complete sentences for generating listening trend conclusions and recommendations are generated using a Hidden Markov Model (HMM).

[0115] The generation of complete sentences by a Hidden Markov Model can be specifically as follows:

[0116] Observation set: Observations are usually words or characters;

[0117] First, based on the hearing trend predicted by the hearing test model, unique text is generated, such as "rising" or "falling". Second, unique text is generated based on the hearing level, such as "normal hearing", "no communication barriers", "severe hearing loss", or "cannot hear sounds". Then, suggestive text is generated based on recent hearing results and trends, such as "seek medical attention promptly" or "keep ear canals clean". Using this method, a vocabulary list {from, several times, recently, test, status... trend, rising, falling, leveling off} is built. Finally, a unique index is assigned to each word or character in the vocabulary list.

[0118] Initial state probability: The probability of each state at the beginning of the model, π = {πi}, where πi = P(si) represents the probability of being in state si at the initial time.

[0119] State transition probability: A = {aij}, where {aij} = P(sj|si) represents the probability of transitioning from state si to state sj.

[0120] Observation probability: B = {bik}, where bik = {ok|si} represents the probability of observation in state si.

[0121] The working process of Hidden Markov Models:

[0122] If we want a complete test conclusion and recommendation, it would be like this: Based on the recent hearing tests, your hearing condition is trending flat with no significant changes. However, you may still have difficulty hearing sounds during conversations, which may cause some inconvenience to your life. We recommend that you seek medical attention as soon as possible. At the same time, you should protect your hearing, pay attention to ear canal cleaning, avoid using sharp objects to clean your ears, and avoid prolonged exposure to high-noise environments.

[0123] In practical applications, the text generation steps are as follows:

[0124] 1. Initialization: Select an initial state s1 from the possible states based on the initial state probability. If the initial state is "your", then calculate the probability of the character after "your" appearing based on the state transition probability, i.e., the state transition probability. According to the given training set, the word with a higher probability after "your" is "listening". At this time, the sentence becomes "your listening", and the state becomes "your listening". The change of state will inevitably affect the observation probability.

[0125] 2. Generate Observations: Based on the current state, generate observations (i.e., a word) using the corresponding observation probability bik. For example, if the state is "Your Hearing", the calculated observation probability of the next word is such that, according to the training set, the word "condition" has a high probability of appearing among the possible words, then the sentence might become "Your Hearing Condition".

[0126] 3. State transition: Based on the current state, the next state is determined using the transition probability aij.

[0127] 4. Repeat steps 2 and 3 until a termination condition is met, using a period as the terminator. This will result in a complete sentence summarizing the user's or patient's hearing assessment conclusions and recommendations based on the analysis of their historical hearing test records.

[0128] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0129] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based intelligent hearing screening system, characterized in that: It includes a hearing test terminal (1) and a back-end server terminal (2); The hearing test terminal (1) detects the patient's / user's hearing data during the test period and saves it to the backend server (2); The backend server (2) includes: Data module (3): Saves real-time hearing data and historical hearing status data of patients / users; Analysis controller (4): Based on the preset hearing data and hearing classification comparison strategy, determine the hearing status of the patient / user based on the real-time saved hearing data; Hearing trend module (5): Based on the analysis controller (4), the hearing status of patients / users is determined in real time and the historical hearing status data of patients / users is analyzed to generate hearing trend results; Hearing test results module (6): Based on the preset hearing data and hearing classification comparison strategy, the analysis controller (4) determines the patient / user's hearing status in real time and the hearing trend module (5) generates hearing trend results data of the patient / user as the basis for judgment to generate hearing suggestions and trend text.

2. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: The hearing measurement terminal (1) includes: Ambient noise detection module (7): Detects ambient noise data of the patient / user's environment during the measurement period; Hearing test module (8): Separates environmental noise data and tests patient / user hearing data within the measurement period.

3. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: The hearing classification in the preset hearing data and hearing classification comparison strategy of the analysis controller (4) is as follows: Level 1: The user's or patient's life is basically unaffected; Level 2: The user's or patient's life will be affected, and communication will be hindered; Level 3: The user or patient's life is severely affected, and communication is extremely difficult; Level 4: The user or patient has no problem with hearing communication, but has difficulty locating the sound source.

4. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: The hearing trend module (5) generates hearing trend results based on the real-time determination of the patient / user's hearing status and the analysis of the patient / user's historical status data by the analysis controller (4). Specifically: Predicting hearing trends using time series models; The time series model function is as follows: f(x)=X1+k(t n -t1) Where f(x) is a time series, X1, X2, ..., X... n The hearing level is graded for each test, k represents the hearing trend, and t1, t2, ..., t3 are the gradations for each test. n For time; Where t1 corresponds to X1, t2 corresponds to X2, ..., tt n Corresponding to X n .

5. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: The hearing trend module (5) generates hearing trend results based on the real-time determination of the patient / user's hearing status and the analysis of the patient / user's historical hearing status data by the analysis controller (4). Specifically: S1. Construct a Gaussian process model; S2, Based on dataset D = {(t n ,X n Training a Gaussian process model; Among them, t n It is the time point for the hearing test, and X n This corresponds to the hearing level; S3. Predicts real-time patient / user hearing trend results based on trained Gaussian process models; S4. Output the predicted hearing trend results.

6. The intelligent hearing screening system based on machine learning according to claim 5, characterized in that: The Gaussian process model function is as follows: § t ~GP(m(t),k(t,s)); Where m(t) is the mean function describing the central tendency of the data points, and k(s,t) is the kernel function defining the correlation between time points s and t, where s and t represent any two time points.

7. The intelligent hearing screening system based on machine learning according to claim 6, characterized in that: The kernel function k(s,t) is a radial basis function (RBF), as follows: Where, σ 2 It is the variance, where s and t are two distinct time points in the continuous domain of the Gaussian process domain, ||st||. 2 It is the distance between two different time points s and t.

8. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: The hearing test result module (6) determines the patient / user's hearing status in real time based on the preset hearing data and hearing classification comparison strategy, and the analysis controller (4). The hearing trend module (5) generates hearing suggestions and trend text based on the patient / user's hearing trend result data, specifically: S1. Construct a hidden Markov model based on hearing data; S2. Train the Hidden Markov Model; S3. Generate real-time patient / user hearing suggestions and trend text based on trained Hidden Markov Models; S4. Output listening suggestions and trend text.

9. The intelligent hearing screening system based on machine learning according to claim 1, characterized in that: Initialization: Select an initial state based on the initial state probability. Generate an observation sequence: Generate an observation based on the current state and the observation probability. State transition: The state transitions to the next state based on the current state and the state transition probability. Repeated generation and transfer: Repeat the above steps until a predetermined termination condition is met, such as generating a specific text length or encountering a terminator. Generate complete sentences of listening trend conclusions and recommendations using a Hidden Markov Model (HMM).

Images